Video Friday is your weekly selection of awesome robotics videos, collected by your friends at IEEE Spectrum robotics. We also post a weekly calendar of upcoming robotics events for the next few months. Please send us your events for inclusion.
Nvidia just paid US $12.9 billion for the company that acquired Pollen Robotics, and this must be why.
Meet Microduck. 🦆 The 25-centimeter, 780-gram robot that waddles, falls, gets back up, and learns new tricks.
Packed inside: 15 degrees of freedom, a front camera, an 8x8 lidar, two IMUs, mics, a speaker, NFC, Wi-Fi, and Bluetooth.
Out of the box, Microduck already walks, sits, crouches, roller skates, picks up objects with its articulated beak, and recovers from falls on its own. Drive it with a game controller, plug-in accessories, and NFC tagged objects, run autonomous behaviors, or gather several Microducks for races and football.
Software fully open source. Ready for whatever you throw at it.
On pre-order for an astonishingly low $399, and ships before Christmas.
Most fish-inspired robots are built for one size and one job, so scaling them up or down usually means starting from scratch. A team of engineers says it has found a way to solve that problem. They’ve unveiled ScaFi, a robot modeled on fish like cod and mackerel.
Martin writes, “We’re a small robotics team in Czechia, Europe, building practical hardware around the Unitree G1. Here’s a short demo of our lightweight gripper picking up a strawberry; the gripper weighs under 200 grams and is designed for simple, sensitive manipulation without adding a complex multifinger hand.
EmoLo brings emotion-inspired expressive locomotion to Open Duck Mini V2, a low-cost, open-source bipedal robot inspired by Disney’s BDX droids. With a single reinforcement learning policy, the robot can generate distinct walking styles associated with different emotional expressions, showing how characterful and expressive whole-body motion can be achieved on an accessible robotic platform.
If it’s possible for a robot with a completely immobile face to look frustrated, this robot absolutely does, starting at three minutes into this video.
Noble Machines deployed its first general-purpose robots to a Fortune Global 500 industrial customer within 18 months of the company’s launch and met its first delivery milestone, made possible by its AI-driven whole-body control and industry-leading end-to-end autonomy.
We’ve reduced the time it takes to go from physical prompt → robot behavior. The faster anyone can teach a robot to do something new, the easier it becomes to scale physical work.
TRON 2 × Wuji Hand 2 handles TCM pharmacy work: picking, weighing, grinding, and packaging. The omnidirectional base frees the hands, while precise gripping and dual-arm force control enable midair operations.
From about 2017, individuals began to truly connect with the initial wave of companion robots. These devices had personality, moved around, joked, and answered when you spoke to them. Most early companion robots, however, were still limited by simple voice-command interactions and narrow functionality. Once the novelty wore off, many ended up sitting unused on shelves. As some of those companies went out of business and turned off their servers, many owners likened it to losing a pet.
What Ollobot describes as “gentle intelligence” is a useful way to think about where the serious work in this category is going. Not toward more powerful assistants, but toward more present ones.
The problem companion robots were trying to solve
Loneliness is not a niche issue. According to one study, nearly one out of three elderly adults resides alone, meaning they do not have daily companions. Research also shows that children whose parents have migrated for work, leaving them in the care of relatives, were 2.5 times more likely to experience loneliness than children whose parents remain with them. Among working adults living alone in urban environments, similar patterns of social isolation emerge, even if they are less visible.
Over the years, technology has time and again attempted to solve this problem via video calls, smart speakers, and messaging apps without much success. Those tools are geared towards communication between people that already have relationships. They do not create presence. They schedule it. That is the gap that a new generation of AI companion robots is being engineered to fill.
Today’s AI robots are different
Today’s companion robots are not just cute and cuddly. They are designed with psychological research, clinical insight and long-term interaction models to be truly useful in real homes.
Three fundamental shifts define the current generation:
From reactive to proactive response. Older robots relied on you speaking to them, but modern robots monitor a room with cameras, microphones, and surroundings sensors to initiate interactions without your input, and they can pick up on your emotions.
From function-oriented to emotion-oriented design. The original pitch for companion robots was about what they could do. The question driving the serious work now is how they make you feel, which is a harder engineering problem and a more honest framing of what the product is actually for.
From standalone hardware to connected ecosystems. Leading brands are creating platforms rather than devices with software included as a built-in layer and remote access from the beginning.
The global AI companion market size was valued at US $36.8 billion in 2025 and is projected to grow from $48 billion in 2026 to $318 billion by 2033, at a compound annual growth rate of 31 percent from 2026 to 2033.
Three household scenarios and interaction models
Ollobot’s advanced AI family companion robot OlloNi SS1 addresses a number of gaps in what existing technology offers.
Elderly individuals living alone. The combination of proactive interaction, fall detection, and persistent presence addresses both safety and companionship without the social overhead of asking family members to check in more frequently.
Children in households where parents work far from home. The SS1 functions as a consistent companion that already knows a child, their preferences, their moods, and their routines. The remote connection features allow parents to stay present without requiring a scheduled call, and the life recording system gives them a passive window into their child’s days that feels less clinical than a monitoring camera.
Single professionals living alone in cities. The SS1 adapts to daily routines, builds up a preference model over time, and provides ambient social presence without demands.
OlloNi SS1 adapts to daily routines over time.Ollobot
What OlloNi SS1 is doing differently?
Ollobot’s goal in building intelligent companion robots is to address the gaps in technology and capability, using innovation not to automate tasks but to fill emotional voids.
Much of the robotics industry has historically pursued human imitation — machines that speak, look, or behave like people. The SS1 is instead designed around familiarity and long-term coexistence rather than realism.
The system integrates multiple subsystems operating in parallel, including visual perception, audio processing, mobility control, and interaction management. It is equipped with a multi-chip AI 4K vision module capable of facial recognition and motion tracking. One small but revealing detail is the inclusion of a physical privacy cover for the camera — a mechanical solution to concerns that software settings alone may not fully resolve.
OlloNi SS1 can actively integrate into family activities, and it can autonomously move closer to capture memorable moments or reposition itself to remain engaged in ongoing interactions.Ollobot
The robot supports advanced mobility across multiple indoor surfaces, including wooden floors, ceramic tiles, and low-pile carpets, with slope climbing capability up to 3.5 degrees. Rather than remaining in a fixed location, it can move naturally throughout the home to stay close to household members as daily activities unfold.
For example, the OlloNi SS1 may greet family members when they arrive home, follow an older adult from the living room to the kitchen while continuing a conversation, remind a child to take a study break after a prolonged period of inactivity, or notice that someone appears unusually quiet and gently check in. During family activities, it can autonomously move closer to capture memorable moments or reposition itself to remain engaged in ongoing interactions.
The robot continues to evolve over time, with over-the-air updates that deliver new features, performance improvements, and AI enhancements
It also incorporates fall detection with optimized accuracy for safety monitoring scenarios. A 6-microphone array enables omnidirectional voice pickup with an effective voice capture range of up to 5 meters, supporting reliable wake-word detection and far-field interaction.
To support continuous companionship, much of the robot’s AI processing takes place directly on the device through its “heart module” architecture, with 16 GB of memory and 64 GB of local storage. This enables the system to retain household memories, recognize familiar faces, and respond with lower latency, making interactions feel more natural even during everyday routines.
Because companion robots are expected to remain available throughout the day rather than only during brief interactions, the SS1 is designed for extended operation, offering up to 12 hours of standby time and around 5 hours of active interaction on a single charge. This allows it to accompany users through meals, conversations, playtime, and other daily activities without frequent interruptions.
To support engaging interactions, much of the robot’s AI processing takes place directly on the device through its “heart module” architecture.Ollobot
Like the relationships it is designed to build, the robot continues to evolve over time. Running on Android OS with over-the-air (OTA) updates, the system continuously receives new features, performance improvements, and AI enhancements, allowing its capabilities to grow alongside the household it serves.
The robot’s behavioral model also improves over time. Rather than reacting to isolated commands, it attempts to establish a baseline understanding of household routines and individuals. Changes in behavior — prolonged quietness, unusual inactivity, or emotional cues — become triggers for interaction.
Presence instead of utility
Several features in the OlloNi SS1 illustrate this emphasis on presence and continuity in its interactions.
The system can identify different household members, including pets, and adapt responses accordingly. Remote communication features allow family members to connect through the device without treating every interaction like a scheduled call. Environmental sensors support contextual reminders tied to weather or room conditions.
Its “2+1” multi-display configuration is also designed around emotional communication. Two circular side displays function as expressive “emotional eyes,” while a separate primary display handles information and structured interaction. The separation allows emotional signaling and functional communication to operate independently, creating more intuitive nonverbal interaction even when no dialogue is taking place.
The robot’s behavioral model improves over time. Rather than reacting to isolated commands, it attempts to establish a baseline understanding of household routines and individuals.Ollobot
The SS1 also includes an automated life-recording system built on facial recognition and behavioral-event detection that can capture moments such as laughter, physical closeness, or group interaction automatically. An integrated AI vlog engine can then organize those moments into edited short-form videos with automated sequencing and soundtrack generation. The design intent is to preserve spontaneous domestic moments without requiring active documentation behavior from users.
An integrated AI vlog engine can organize recorded moments into edited short-form videos with automated sequencing and soundtrack generation
Visual data is processed primarily on the device through the SS1’s on-device AI architecture, with household memories stored locally and managed within Ollobot’s proprietary ecosystem instead of being shared with third-party smart home platforms. Access to recordings and live feeds is restricted to authorized users through the companion app, while encrypted communication helps protect data during remote access. Users also retain direct control over recording preferences, and the physical camera privacy cover provides an additional hardware-level safeguard whenever visual monitoring is not desired.
Remote communication is similarly structured around persistence rather than transaction. Traditional video calls are episodic and screen-bound; the SS1 instead acts as a continuously present interface embedded inside the household environment. Through autonomous mobility, environmental awareness, and persistent household memory, remote family members interact with an ongoing domestic context.
The larger shift to “gentle intelligence”
Ultimately, gentle intelligence is not about making robots behave more like humans — it is about helping them fit more naturally into human lives. Each OlloNi SS1 unit develops a unique behavioral profile based on its household. Two units running in different homes for a year will have become meaningfully different from each other, shaped by the specific people, habits, and rhythms of where they live.
That kind of long-term personalization is what early companion robots never had. It is also what makes the difference between a product that ends up on a shelf and one that actually earns its place in a home.
Video Friday is your weekly selection of awesome robotics videos, collected by your friends at IEEE Spectrum robotics. We also post a weekly calendar of upcoming robotics events for the next few months. Please send us your events for inclusion.
This is very, very cool. But I’m trying to think of what the commercial use case will be, you know? I guess, high speed, incredibly dangerous package delivery to second-floor windows or something...?
Humans have a remarkable ability to perform new physical skills from only one or a few examples. Our latest robot foundation model, GEN-1.5, exhibits the beginnings of that same ability: It can learn a new task in seconds, from a single example, without gradient updates or fine-tuning. It displays broad capabilities across one-shot and few-shots learning from demonstration, as well as zero-shot physical generalization. Although the tasks are simple and short-horizon, this is the first model we know for which one-shot and few-shots learning of physical skills have emerged at scale. We view these results as a significant step toward our mission of building general intelligence for the physical world.
I will make the cautionary point that for many of these “the model figured it out” tasks, the blog post can only say that there was no relevant pretraining data “to the best of our knowledge.”
As a professional bagpiper who definitely pays very close attention to whatever that annoying tapping noise is coming from the back of the band, I can attest to this group of robot drummers being absolutely top-notch.
What does it take for an aerial robot to move through a sequence of arbitrary poses—fast, precisely, and continuously? Rather than teaching the robot a behavior from data, we asked how far a first-principles analytical model could take us. Through a collaboration between the AIMS Group at the Hong Kong Polytechnic University and DRAGON Lab at the University of Tokyo, we developed the first sequential-convex-programming-based trajectory-optimization framework for generalized multirotors, covering both conventional and omnidirectional platforms.
Gravis Robotics brings robotic intelligence to heavy construction machines. Our retrofit kit, the Gravis Rack, turns off-the-shelf hydraulic machines into robots. Cameras, lidar, and onboard compute lets your machine see and understand its surroundings, and learning-based control lets it work close to its limits, moving more dirt with full, fast cycles.
Robust brachiation requires precise hand movements to grasp and release bars together with highly coordinated whole-body motion. To address this challenge, we propose a learning-based framework centered on waypoint-guided reinforcement learning (WGRL). WGRL guides the end effector through waypoints while allowing RL to explore and generate dynamic whole-body behaviors. With this approach, the learned policy demonstrated robust brachiation across diverse courses with different bar heights, spacings, and orientations in sim-to-sim experiments. In the real world, our life-size dual-arm robot successfully traversed four consecutive bars.
This work introduces a novel approach to physical human-robot interaction (pHRI) by leveraging the joint torque sensors of standard collaborative robots. By mounting a passive, uninstrumented plexiglass touchpad to the robot’s flange, we transform the robot into a handwriting-based input interface.
Microspines are one of many ways to enable robots to latch onto surfaces like walls and ceilings. Now roboticists in Canada are using the mini spikes to get drones to land on a more challenging, remote surface: icebergs.
Like a spider, the Ice Dart can land on and latch onto steep, slippery surfaces such as icebergs and glaciers—an increasingly useful capability as activity in the Arctic increases. The drone can grip onto icy slopes of nearly 60 degrees, which is way beyond what most humans could manage without special equipment.
In a recent study, researchers explained how they developed the Ice Dart drone with a special landing gear that absorbs the impact of a hard landing while holding the drone in place with tiny spines that penetrate and grip the ice.
Published in IEEE Transactions on Field Robotics, the study describes how the Ice Dart was able to land on icebergs and a glacier in southeast Iceland. Tests took place amid persistent winds and temperatures of 0 to 10 °C along the ruggedly breathtaking Fjallsjökull (pronounced “FYATLS-yuh-kuutl”) glacier, which empties into a lagoon filled with icebergs. The drone was able to successfully perch at speeds of up to 3 meters per second and slopes of up to 58 degrees, with a success rate of 100 percent even in wind speeds of 30 km/h.
The researchers were motivated by a desire to allow drones to land almost anywhere in the world, since the availability of safe landing sites is one of the primary limitations on where and how drones can operate. The researchers already have a history of developing drones that can land on fast-moving trucks as well as trailers, boats, and steep roofs.
Ice Perching
“The ability to land rather than hover can fundamentally change how drones are used in the field,” says Alexis Lussier Desbiens, a professor of engineering at Université de Sherbrooke, in Sherbrooke, Quebec, Canada, who coauthored the study. “Once a drone has landed, energy consumption drops dramatically, allowing much longer observation periods with a small aircraft. The drone also becomes completely silent and can even reduce or eliminate its thermal and RF signature by shutting down major onboard systems.”
Landing on icebergs specifically allows drones to monitor them for days or months, producing more detailed observation than a quick aerial surveillance mission. This could simplify iceberg tracking compared to methods such as helicopter deployment, dropped instruments, or dart-like tracking devices, and provide another data layer to satellite and ship-based iceberg detection, according to the researchers. It could also provide a means of monitoring icebergs that are otherwise untrackable.
With its carbon-fiber construction, the Ice Dart drone weighs just 2.65 kg and has four legs arranged in an X shape, attached to its body with a pivot joint. Used in the group’s previous drone research, this landing gear disperses energy to reduce impact and overcomes multiple engineering challenges. The friction shock absorbers consist of 38 disks that generate friction torque as the legs move up and down upon impact. This lowers the UAV’s center of mass and helps spread out the kinetic energy of landing, but the real trick comes in the form of two retractable spines on each foot—one for uphill and one for downhill grip. The larger spine engages on the more heavily loaded downhill feet, and the smaller, thinner spine engages more easily on the uphill feet, even under very low loads on steep slopes. The spines only penetrate the ice as the suspension compresses, generating grip and protecting them from high-impact forces.
“The inspiration for the retractable spines in the feet came from looking at a cat’s claws and their ability to deploy only when needed,” says Isaac Tunney, a Université de Sherbrooke postdoc in mechanical and robotics engineering who was lead author of the paper. “I wanted to create feet that would naturally and passively engage their spines in the ice at the right moment, regardless of the drone’s orientation, the surface geometry, or the ice conditions.”
Arctic Surveillance
William D. Harcourt is a researcher at the University of Aberdeen, in Aberdeen, Scotland, focused on Arctic glaciers, snow, and sea ice, as well as the use of remote sensing and machine learning techniques. Harcourt was not involved in the study, but he sees several potentially interesting applications of the technology.
“Near the front of tidewater glaciers, these systems could enable measurement of stress and strain and help us understand calving processes,” Harcourt says. “Drones can be used as a mobile GPS, literally acting as a receiver on the ice, but the system would need to solve tilting issues as 3D change measurements usually required the antenna to be horizontal. However, if these problems can be solved, it could be used to track iceberg movements.”
The researchers want to continue developing the Ice Dart technology for real-world applications, including autonomous landing site selection and an emergency takeoff capability to be used if an iceberg rolls over or breaks apart. This August, the drone will be deployed during a Canadian Arctic mission to land on icebergs, collect data, and help validate ship-based iceberg-detection systems.
Humanoids desperately need to stop making YouTube videos and get a job already, and Persona AI is one of the few humanoid companies that seems to be entirely focused on making that happen. Persona AI’s approach has been to carefully select a job that is economically viable for robots right now, and they’ve found one that was also the job of one of the very first industrial robots ever sold: welding.
As of our first conversation in 2024, Persona had committed to building an economically viable humanoid, but they hadn’t yet figured out where their focus was going to be. “We were all over the place,” Radford says. “Warehousing, automotive, we probably even mentioned the home.” These are the same environments with the same sorts of potential applications that basically every other humanoid robotics company is attempting to make economically viable, and despite an ever more exhaustive number of demonstrations, so far none have succeeded at any sort of useful scale.
The challenge for Persona, and really for every robotics company, is that it’s not enough that you have a robot that is simply capable of doing a task. It’s also not enough that your robot can do that task in a way that is efficient, reliable, and safe. What’s required is that your robot can make money for both you and your customer. Most humanoid companies seek to achieve this by targeting baseline “unskilled” human labor.
Persona did not see economic viability in the unskilled labor approach, Radford says. “We started forming this thesis around skilled trades and tool usage.” Persona is targeting much more expensive skilled labor with its robots, and the reason why this is feasible is because their entry point focuses on the kind of skills that robots are especially good at. “I like to call it ‘last-mover advantage.’ We’ve seen everything that everybody’s doing, and we’ve decided that there’s a different way.”
The first task that Persona’s humanoid is focusing on is welding—using a handheld tool to connect one piece of metal to another. “Tool use is pretty difficult,” Pratt says. “And we want to use the same tools that humans do, which makes it more difficult.” That difficulty is offset somewhat by the fact that Persona’s humanoid will first focus on making long, linear welds that are relatively uncomplicated. “This is not the hardest style of weld,” Radford says, “but in shipbuilding you need a lot of them—hundreds of kilometers of linear welds per ship.”
Currently, Persona has two public partnerships: one with HD Hyundai, which is the world’s largest shipbuilder, and the other with POSCO, one of the largest steel producers in the world, both in Korea. Persona declined to get into detail, but Radford says that broadly speaking, the company is interested in customers who can support ‘hundreds’ of robots per location.
Shipyard welding is an enticing application for Persona because there is a deficit of skilled (and highly paid) workers, it’s taxing physical labor, and it’s a comparatively easy skill for a humanoid to learn.
The welding process is skilled in a very robot-friendly way. Because you can only weld as fast as metal melts, the top speed for the task is an easily manageable centimeter per second. And making a high quality weld involves millimeter-scale repeated motions, which robots excel at, especially over long periods of time—whereas humans tend to get tired or bored. Pratt expects that for these uncomplicated welds, performing on par with humans—if not eventually better—will be achievable soon.
Shipyards make a compelling case for a humanoid with legs, as opposed to a more stable wheeled base. “These open-air shipyards are a couple hundred meters long, with horizontal and vertical spars that you have to step over all the time,” Radford says. “You’ve got to work on the ground, overhead, and through portholes.” Persona considered other form factors, like four legs (or even more), but determined a two-legged robot would be the least disruptive to existing shipyard rhythms.
The Economic Viability of Humanoids
Deploying their robots in shipyards specifically brings additional advantages for Persona. The safety concerns that come with bipedal robots—such as potentially falling over on a human worker—are lessened because a shipyard environment is staffed with workers who are trained to work around potentially dangerous industrial equipment. The company is also less sensitive to competitive pricing because no other company is pursuing the use case. “We’re now in an industry where the value added by our robot can be high enough that we don’t have to cut corners on quality and features in order to reduce the price,” Pratt says. “With a robot for the home, for example, there would be a lot of competition and a ton of price pressure.”
The added value for shipbuilders, Radford explains, doesn’t come from replacing humans with robots. “Our current partnerships are running at a significant backlog, and they’re labor-constrained. So we want to help our customers’ top line, not necessarily their bottom line.” In other words, rather than trying to argue that their robots will lower shipbuilding costs, Persona is instead arguing that their robots will allow more ships to be built. “Even if our robot was more expensive than a human, that would still be valuable to these companies, because it could unlock additional revenue,” Radford says. And when the additional skilled labor does not exist, Persona’s robots could be the next-best option for shipbuilders who need to scale.
Shipyards are environments where legs are necessary for a robot to be useful.CFOTO/Future Publishing/Getty Images
Persona’s Multipurpose Future
In the current commercial humanoid climate, where the emphasis seems to be on developing a “general purpose” robot (whatever that means) that will somehow justify itself through some undefined scale in some equally undefined and perpetually receding future, Persona stands out with their focus on a seemingly viable, near-term, and very specific business case. It hasn’t been easy, though. “It hurts us a little bit,” Radford says. “We’ve been told that we’re not thinking big enough.”
But a tool-using heavy industrial humanoid has plenty of future applications, many of which can be expanded from the welding skill even within shipyards. “Shipbuilding is a great beachhead,” Pratt says. “There are tons of adjacent markets, like grinding, painting, and other kinds of fabrication.” Persona’s ambition, Radford adds, is to be “the largest repository of industrial skills.”
It’s going to take time to get there. That time will be needed to collect tens of thousands of hours of expert demonstration data, create high-fidelity simulations, and conduct real-world testing. And however promising Persona’s approach may seem, the company still has to prove that its idea for an economically viable robotics company can be realized. It’s the same challenge that every humanoid robot company is facing. “A lot of the technical problems are the same no matter whether you’re in a house or a shipyard,” Pratt says. “Everybody’s got a great team and smart people, and we’re all knocking these problems out together.”
Video Friday is your weekly selection of awesome robotics videos, collected by your friends at IEEE Spectrum robotics. We also post a weekly calendar of upcoming robotics events for the next few months. Please send us your events for inclusion.
Speaking from experience, I can tell you that the best part of any DARPA challenge is when things go horribly wrong. And after you enjoy all the crashes (followed by all of the battery fires), get caught up with the DARPA Lift Challenge with video recaps of the final few days.
This tiny little robot called STEMbot (as in stem, not STEM) can climb up and around plant stems to check for pests. It’s not very fast, but it sure is adorable.
How can a legged robot organize its own walking while maintaining a desired direction? In this work, we present a Differential Adaptive Steering (DAST) mechanism for directional adaptation in legged robots under decentralized adaptive control.
If we as a society would just accept that the appropriate place to store clothing is in a pile on the floor, robots would have a much easier time of it.
Objects constructed by robots are ubiquitous. If you’ve used a car, household appliance, or smartphone today, you’ve used an object constructed at least in part by robots. The more products that manufacturers want to produce (and consumers want to consume) at lower costs, the more industrial robots will be needed.
There are over 4 million industrial robots in use worldwide, according to the International Federation of Robotics. And researchers predict that number will grow to over 16 million by 2030, as manufacturing rapidly increases. But what’s going to happen when they start breaking down? A new system designed by researchers at the Karlsruhe Institute of Technology (KIT), in Karlsruhe, Germany, can predict the defect in a broken product and disassemble it while protecting valuable parts from damage. To continue robotic development sustainably, the industry should prepare for the dismantling, recycling, and rebuilding of our robotic systems.
The system consists of a predictive algorithm that guesses how a product is broken, along with robotic manipulators that actually take the broken product apart. At every stage of the process, the system checks to see if the results align with its predictions, and updates its methods if necessary. For example, in the video below, the system begins by unscrewing a broken component. To simulate a stuck screw, the researcher replaces the screw. When the system observes the screw still in place, it switches to milling away material to remove the part.
Building a product with new parts is easy, says Jan Baumgärtner, one of the designers of the system. Each step is clearly outlined, and there are no expected deviations. But taking apart something that’s broken is unpredictable. “We can imagine 100 ways that something can go wrong.” And if you start taking something apart without knowing how it broke, you might have to undo part of your work when you find the problem. For example, if you have to unscrew 100 screws holding two parts together, but the last screw is stuck, you’ll have wasted time unscrewing all those screws when you should have used a different method to remove the part in the first place.
How to Take Apart a Product
KIT’s robotic disassembly system relies on a CAD model of the broken product and of each part, so it can see how the parts should behave and understand if anything is out of the ordinary. It also uses a mathematical model to predict the damage done to a broken part.
When you give the system a broken device and a CAD model, it first guesses how each part of the broken device should move. The axes each part can move along are called degrees of freedom (for example, a screw should rotate, but not move side to side). The disassembler nudges each part to see if it moves as expected. Based on how the part actually moves, it then uses the mathematical model to predict what went wrong with the part: A corroded part might move less than you think it should, a loose screw may move more, and a deformed part might have different degrees of freedom than expected.
At the beginning of disassembly, the system formulates a plan. It guesses what might be wrong with the device it’s taking apart, and then can change its guess based on observing each piece it takes apart. For example, if there was a screw loose in the part, that might be hard to guess from an initial photograph of the broken part. But when the system moves the screw, it will notice that it can move in more ways than a screw should move, and take that loose screw into account when deconstructing the device. You can also tell the disassembly system which parts are most important to salvage intact from a broken device, and it can adjust its strategy to preserve those specific parts.
The Automated Circular Economy
Baumgärtner’s motivation behind the design of the robotic disassembler is to help create a circular economy, where old devices are repaired instead of thrown away, reducing waste. “The big future is saving our planet,” he says.
Baumgärtner envisions scaling up this one system, composed of a few robotic arms, to have many robotic disassembler arms, each with different tools. These arms will specialize in a different part of the disassembly process so that an entire factory could use different robotic limbs to disassemble a wide range of products. Think of an industrial robot factory that creates cars, but instead is specialized to take them apart. Or, as he puts it, “as a giant robot with 100 arms.”
Ultimately, if this system works as intended, it would be a fully automated way of extracting a broken part from a system, replacing it, and rebuilding the device. Then the circular economy would really shine, as people replaced broken parts in old devices instead of buying new ones all the time. “That’s why we need to think about scaling this,” he says. “Because it means it becomes so cheap that it’s cheaper to repair this [electronic device] than to produce it. That’s the goal.”
This research was presented at the IEEE International Conference on Robotics and Automation (ICRA) 2026 in Vienna.
This story was updated 11 August 2026 to clarify that the disassembly system works for products in general, not only robots.
Video Friday is your weekly selection of awesome robotics videos, collected by your friends at IEEE Spectrum robotics. We also post a weekly calendar of upcoming robotics events for the next few months. Please send us your events for inclusion.
The DARPA Lift Challenge is taking place through this weekend. There are a couple of very brief overview videos from the past couple of days, which are only really interesting because they give you a quick look at some utterly bizarre heavy-lift drone designs. If you like what you see, DARPA has recorded livestreams of the entire event so far. We’ve posted one of those at the end of this section, and if you want to be impressed by some super-weird drones, check out this and this.
When NASA’s SkyFall helicopters take to the Martian skies, one of their tasks will be to hunt for frozen water—a critical resource for future astronauts—using ground-penetrating radar. For that radar to work, the rotorcraft will carry a flexible, fabric-based antenna that extends below the aircraft without interfering with landings or breaking at touchdown.
We’ve improved how GEN-1 learns to adapt to new actuators and new robots at the lowest level, with up to 10-20x gains on internal benchmarks. This significantly boosts performance on high-precision tasks like disassembling parts from a NIST board.
We present SonicFly, a passive aeroacoustic perception framework that enables one unmanned aerial vehicle (UAV) to estimate and follow another using only the leader’s intrinsic flight sound.
The U.S. Federal Communications Commission (FCC) “Covered List,” originally published in 2021, identifies communications equipment and services that it says pose a threat to national security. On 28 July, the FCC added mobile, communicating robots weighing more than 2 kilograms and power inverters commonly used in solar panels to the list, meaning that new products from any foreign country in these categories are no longer eligible for import.
The move is a Department of Defense–driven expansion of scattered federal efforts to further limit U.S. exposure to potentially sensitive Chinese technology, but it may impose major changes on the robotics industry in allied countries, too.
All foreign-produced advanced robotic devices pose an unacceptable risk to the national security of the United States and to the safety and security of U.S. persons…unless the [Department of Defense determines that] a given foreign-produced advanced robotic device, or a class of such devices, does not pose such risks.
There are two important definitions here. The first is what an “advanced robotic device” is, and the second is what “unacceptable risk” means. Drones already went through their own round of this sort of regulation, so they’re exempt from this particular restriction, as are connected vehicles and medical devices. As far as the FCC is concerned, “advanced robotic devices” are mobile systems that incorporate on-board sensing and communications and have some amount of autonomy. There are a couple of loopholes, including systems weighing under 2 kilograms and any system that communicates at less than 200 kilobits per second, which opens up some creative possibilities. It’s important to note that this applies to new devices; those already certified are not restricted for sale or use.
As to the risks, the U.S. government says that foreign advanced robotic devices represent “a cybersecurity risk that threatens the security of critical infrastructure and thus the safety and security of U.S. persons.” There seem to be two main points to the justification, found in Appendix C. The first is that mobile robots are important to both the economy and the military, so the United States needs its own supply chain and industrial base rather than relying on foreign manufacturers. And second, mobile robots monitor critical infrastructure in sensitive locations, making them a security risk.
The Country That Must Not Be Named
As part of its justification for why foreign robots are a security risk, the DOD cites IEEE Spectrum’s article on a critical vulnerability in robots from Unitree, based in Hangzhou, China, along with several other news articles and reports about Chinese robotics. And despite the FCC swearing up and down that this action is “country neutral” and “not targeted at any country or countries,” U.S. national security sources told Spectrum that the perceived threat is obviously China. That’s how China feels about it, too, per a Chinese Ministry of Commerce 29 July press conference (translation of the first quote here):
On the surface, the FCC’s measures fly the banner of “non-discrimination,” but in substance they discriminate against and suppress Chinese enterprises and products…
China firmly opposes the U.S. overstretching the concept of national security and going after Chinese companies. Protectionism does not make the U.S. more competitive and will only hurt the interests of U.S. companies and consumers. China will continue to do what is necessary to firmly defend the legitimate and lawful rights and interests of Chinese companies.
It’s unclear what China is going to do about this—but how about the rest of the world? How can foreign companies that make advanced robotic devices get them cleared for FCC authorization? Among many, many other things, you’ll need to provide “a detailed, time-bound plan to establish or expand manufacturing in the United States for the advanced robotic device.”
Because China also produces a large fraction of robot components, even for robots assembled in the United States, it will have strong leverage in any related negotiations until U.S. robotics companies further diversify their supply chains.
Applicants must also submit their applications to the DOD and FCC by 1 January 2028, which is unfortunate for anyone who wants to develop an advanced robotic device after that point.
Robotics Industry Reactions
This is all very new, and reactions from the robotics community have been mixed.
Some American robotics companies may benefit in the local market from the newfound lack of competition in the commercial market. Brendan Schulman,Boston Dynamics’ vice president of policy, wrote an enthusiastic endorsement of the ban on LinkedIn: “I sense that this is just the first round in a series of policies that will define the success and growth of the industry for decades to come.” On the other hand, third-country buyers may just stick to Chinese products, as they generally have for drones and electric cars.
But not all companies expect major changes from the new regulation. American customers “need to know they can audit the technology, get support quickly, and keep the system operating without depending on a fragile overseas supply chain,” Nic Radford, the CEO of the U.S. humanoid robotics company Persona, tells IEEE Spectrum. In other words, he figures some customers wouldn’t have wanted Chinese humanoids anyway.
Philipp Frey, vice president of strategy for the Swiss quadruped company ANYbotics, agrees. He says their enterprise customers in the United States “increasingly evaluate robots on long-term reliability, cybersecurity, software capability, safety certification, serviceability, and ecosystem integration, not on hardware cost alone.”
ANYbotics also plans to apply for conditional approval of future products, Frey says. That will involve a national-security review by the DOD or the Department of Homeland Security, disclosing company beneficial ownership, supply-chain risks, and declaring a plan for establishing a significant manufacturing presence in the United States.
Gavin Kenneally, CEO of the U.S. quadruped company Ghost Robotics, is more explicit about the risks that Chinese robot strategy poses to the United States. “Active and purposeful spyware is deployed inside the U.S. on Chinese robots. Examples of predatory pricing abound. And this isn’t just a competition between U.S. and Chinese robotics companies; it’s between private U.S. companies and China’s coordinated national strategy,” Kenneally tells Spectrum. “If today’s announcement encourages stronger cybersecurity and a more level competitive environment, that’s good for customers and good for the robotics industry.”
So is an industry-wide ban the best way to guard against threats? American approaches to Chinese technology security risks have been “ad hoc and fragmented,” wrote the Brookings Institution sociologist Kyle Chan in a report published 9 July. Chan called for the Bureau of Industry and Security, part of the Department of Commerce, to centralize federal information gathering and decision-making on how to handle risky foreign devices. He also called for better public input mechanisms for these issues, and a continuous, proportionate process that tightened or relaxed targeted import restrictions in response to well-defined risks.
That would allow American industry to continue benefiting from partnerships with Chinese manufacturers in less sensitive links of the supply chain, Chan argues. Those links will evolve over time, requiring continued assessment, but without those partnerships, crude bans “could make it more difficult for American startups and researchers to develop new software and end up slowing innovation across the U.S. robotics ecosystem,” he writes.
For a while there, it seemed as though robotics as a whole was stuck in a mad rush towards building humanoid robots mostly because it was very possible (and very lucrative) to do so, even without near-term goals that were necessarily realistic. Some of the magic of those first couple of years of the humanoid explosion has stuck around, but there’s also been an industry-wide sobering leading to pointed questions about practicality and value. In other words, starting a commercial humanoid company now is a much different proposition than it would have been just a few years ago.
On 15 July, Walden Robotics emerged from stealth with US $300 million in funding at a valuation of $1.1 billion. Walden is a spinout of Toyota Research Institute (TRI), and it’s spent the last 10 or so years working on hard problems in robotics with the goal of transitioning from research to real-world applications. That seems like the amount of time and experience that it might reasonably take to develop a practical and value-driven approach to deploying general-purpose humanoid robots, and Walden has chosen an excellent starting point by skipping the legs.
“It’s ironic,” says Walden cofounder and CEO Russ Tedrake. “I thought about legs for 20 years; that’s the class I teach at MIT. There are many reasons to build a robot with legs. But the question is, what’s the addressable market? And what percentage of it is covered by a wheeled base?” It’s this focused, practical thinking that sets Walden somewhat apart from many (if not most) of the other companies in this space. Rather than developing a robot first and searching for a viable commercial use case second, Walden instead identified applications where robots can provide value now, and designed a robot that could safely and efficiently meet those needs.
Walden Robotics
Walden Robotics’ Manufacturing Focus
Russ Tedrake is the CEO and cofounder of Walden Robotics.Walden Robotics
Tedrake is light on the details about what specific applications Walden is targeting at this point (citing confidentiality with current commercial partners). Manufacturing and logistics environments where there are a lot of relatively simple and repetitive tasks that aren’t friendly to conveyor belts and preprogrammed robot arms are a good bet. Even in these environments, however, robots still have to find a useful niche because they’re going up against human workers who are more flexible while also cheaper to employ. So the question is: How do you make an argument to a customer that a robot is actually a better solution than their existing human workers?
“You need to find applications with high utilization—where the robot is used 24 hours a day, 7 days a week,” says Tedrake. “Manufacturing is a global imperative right now, and it makes the economics work.” Economic viability is a necessary condition, but it’s not a sufficient one for Walden, or for their partnership with Toyota. People are a big part of Walden’s plan, too.
One of Walden’s major strengths is the company’s partnership with Toyota, which is not all that surprising given that Walden is a spinout from TRI, which is Toyota’s Silicon Valley–based R&D arm. “Toyota was very proud of the work we had done at TRI, and was ready to go big in this space,” says Tedrake. “Part of the excitement of having Toyota as a partner is that their culture is deeply people-first. When talking to Toyota’s leadership, I was never asked how much money this is going to make, but I was asked how it will improve the quality of life for all people.”
The robot’s chonky design allows it to meet the high-payload requirements of useful manufacturing work.Walden Robotics
In this context, at least in the short term, Walden’s approach to improving the quality of life for people is to take over those aforementioned repetitive manufacturing tasks with robots. Tedrake hopes that this will lead to workplaces where skilled craftspeople are able to do even more with their hard-earned expertise, increasing their efficiency, productivity, and happiness all at the same time—a noble goal, although there’s only so much Walden itself can do to make this happen, and not all customers will share Toyota’s priorities.
Wheeled Humanoid Robots in Factories
Many other humanoidroboticscompanies are also targeting these logistics and manufacturing spaces with general-purpose robots, and they’re doing so by making robots that are as humanlike as possible. The theory is that a humanoid form factor is necessary when operating in human environments. And there are certainly arguments in favor of a humanoid with legs—stairs exist, for one, and legged robots have a smaller footprint compared with ones that have wheels.
But a large wheeled base offers some significant advantages, as Tedrake points out. You’re incentivized to cram the base full of batteries, since more weight near the floor keeps the robot stable, which also solves the problem of running out of power during the middle of the workday. More importantly, a statically stable robot that moves around on a wheeled base can bypass the safety challenges that are currently keeping legged humanoids physically separated from real humans—most prominently, the fact that legged robots can fall over. “Factories already have autonomous mobile [wheeled] robots,” explains Tedrake. “They already have safety cases built around AMRs. You can piggyback on that with a wheeled base.”
Simple, rugged grippers make the robot suitable for commercial deployment.Walden Robotics
Walden’s perspective on manipulation is similar. Many humanoid companies are using five-fingered hands that are highly dexterous but also highly complex, which Tedrake believes is not a pragmatic approach in the context of commercial deployments. “There’s a question of what you need to do the tasks, but the real question is just durability,” Tedrake says. “We have been deployed in a Toyota factory, and at the end of the week, the hands take a beating, so we built hands that can take that. I have not seen a more dexterous hand that could have done the work our hand has done.”
Walden’s long-term plan is to build “general-purpose robots.” It’s not always clear what a general-purpose robot is, because (I would argue) nobody is quite sure what “general purpose” means. It’s certainly not referring to robots that can do everything; I think the closest we can get are robots that can be taught to do a useful number of different skills, which is why I prefer the term “multipurpose.” It’s a little pedantic, I know, but I think the distinction is important because it moderates expectations in the near term.
Part of where Walden’s optimism towards general purposeness comes from is TRI’s earlier research on diffusion policy, which helps robots learn new skills more quickly by leveraging previously learned skills as a foundation. “Fundamentally, multitasking is a way to get to a general-purpose robot,” Tedrake says. “I believe there is a single platform that can do a lot of tasks that are of high value for real customers. That will give us the experience we need to give birth to this deployed general-purpose capability.”
Video Friday is your weekly selection of awesome robotics videos, collected by your friends at IEEE Spectrum robotics. We also post a weekly calendar of upcoming robotics events for the next few months. Please send us your events for inclusion.
Introducing Gemini Robotics 2—the intelligence layer powering the next generation of truly adaptable robots. As it takes its first literal steps, this major advance unlocks intelligent whole-body control, advanced dexterity, and multirobot collaboration.
Humans routinely communicate through abstractions of their bodies, including shadows, silhouettes, and reflections. Here, we present a robotic system capable of dynamic shadow expression using a 21-degrees-of-freedom dexterous hand with compliant soft skin and a learned shadow self-model.
Imagine running your fingertip over the surface of a U.S. penny. You would feel the ridges of the raised letters and numbers, Abe Lincoln’s bearded side profile, and, if it’s tails, the fluted columns of the Lincoln Memorial. Getting a robot to sense the same things is an imposing task, often requiring gathering data on pressure and force at many spatial locations at once. But that’s just what a team of scientists in Europe has now managed to do, using an unusual, colorful robotic skin that provides high-resolution sensing in real time.
“To be honest, when they showed us this, we thought it was, and pardon my French, [expletive] cool, because it’s a distinctly different approach,” recalled Rich Walker, director of Shadow Robot, the U.K.’s longest-running robot company, which primarily focuses on robotic hands.
The research team, which hails from Queen Mary University of London, the University of Florence, the University of Trieste, and the University of Trento, designed a robotic fingertip with a synthetic skin that reflects different colors of light in response to mechanical deformation. By reading the light reflected off the skin, the fingertip generates maps of topology, strain, and contact pressure. The team has already used the sensor to generate maps of a human fingertip, a penny, and a leaf.
Giacomo Sasso, a postdoctoral research associate in the lab of Federico Carpi at Queen Mary University of London, came up with the idea for the sensor. He had been researching optics when he stumbled upon an interesting paper published in the journal Nature. It described the “mechanochromic material” that would eventually make up the reflector in the skin.
Following the method described in Nature, Sasso exposed a light-sensitive film to a 5-megawatt, 635-nanometer (red) laser for seven minutes. The laser beam creates an interference pattern which causes the film to polymerize in alternating densities, creating layers with different refractive indices.
This structure is called a Bragg reflector. The alternating densities and refractive indices in the polymer cause specific wavelengths of light to be reflected. When the reflector is deformed by contact with an object, its layers are stretched, becoming thinner and reflecting light of a different wavelength.
It took Sasso less than a week to re-create the material in the lab. “From there, we started seeing how we could translate these color patterns into something that was useful for us,” he says. Soon, they realized that the color produced by the material was all they needed to be able to sense the topology of objects.
In the robotic finger, the Bragg reflector is sandwiched between a layer of silicone, which protects it from the outside, and a transparent, fingertip-shaped silicone finger with a camera and LED light embedded inside of it.
The light from the LED shines through the clear polymer of the finger. When the fingertip is deformed by an object, the reflector bounces light back to the camera, with wavelengths depending on the level of deformation—red for least deformation, shifting to green, and then to blue when most deformed.
The team also made adjustments to increase the sensitivity of the skin and help the camera to better read color differences. The silicone of the outer layer of the fingertip is colored black to increase the color contrast, allowing the camera to better translate color into the morphology. The rigidity of the camera inside the finger also enhances the deformation of the reflector, producing greater differences in reflected wavelengths.
After all that optimization, the finger provided 100-micrometer resolution with no computational latency, the researchers determined.
What robot fingers need
Human skin takes in a variety of tactile information in order to successfully move and manipulate objects, including temperature, texture, pressure, and vibration. But engineering a robot to do the same is challenging because of spatial constraints. There often isn’t enough room in a robotic fingertip to incorporate more than one type of sensor. The question then becomes: Which type of sensor should be used?
“And the answer to that is…that’s a really hard question. No one knows yet,” says Walker. Carpi’s team’s robotic finger is exciting because it presents yet another option for roboticists to experiment with, Walker says.
Although Carpi’s team isn’t the first to use soft materials for tactile sensing, its technology is unique because it is able to extract quantitative information about depth and size from the topologic maps it generates. According to Walker, most sensors can only generate topological maps, which reveal the relative sizes of object features.
To Sasso, another key advantage of this robotic finger is that it embeds tactile sensing directly into the material of the finger, rather than using taxels, or pixels that measure force or pressure at specific spatial points.
“The core aspect of the sensor is that we’re essentially [moving toward] having the sensing element at the material level,” he says. “The camera, which is a very highly optimized electronic component, is translating whatever the material is already doing directly into digital signals.”
Michael Wang, co-founder and chief scientist at Daimon Robotics, which, unlike Shadow Robot, primarily uses vision-based sensing, echoed Walker’s sentiment that it’s beneficial to explore new methods of sensing, which may bring unique advantages. But he also explained that soft materials often face challenges with durability, and that the significance of the team’s work would be revealed when the finger is integrated into real robot hands.
“The practical and useful benefits, especially in the context of robot hands, remain to be tested and validated,” he says.
When the materials of soft sensors, like the silicone in Carpi’s team’s fingertip, become eroded or damaged after repeated use, the signals measured by the sensors may not reflect objects’ topography as well.
“Especially if you have the electronics embedded into the material layer itself, that becomes a very challenging engineering problem. And I haven’t seen [many] good soft electronics materials that really can undergo long periods of usage,” Wang says.
However, because the Bragg reflector isn’t in direct contact with objects itself, the outer layer of silicone material acts as a protective barrier, Sasso says. The silicone can also be made more durable using certain chemical coatings, according to Wang.
The team has already been talking to companies that could potentially employ the new sensor. They also hope to improve the sensor so that it can sense objects that don’t lie flat on surfaces. That could open up its use in surgical instruments that require precise contact mapping of tissues and organs, Carpi says.
“There are significant developments that we expect with a clear path toward transition to real world applications,” he says.
Atop a lab bench, Cornell Tech postdoctoral researcher Yifan He positions the lens of an optical receiver almost a meter away from an LED emitting a beam of red light. The computer monitor attached to the receiver takes a beat to refresh, then displays an array of squares that resemble a QR code.
When you hold your phone camera up to a QR code, light strikes the image sensor as only a first step to revealing the data hidden behind the black-and-white matrix. The receiver here is doing something different: directly altering its own memory using the photocurrents produced by the beamed array of light. And unlike the data behind a QR code, which might point to a simple web address, this optical code could convey the parameters of an AI model.
The new receiver design, presented last month at the IEEE/JSAP Symposium on VLSI Technology & Circuits in Honolulu, seeks to reduce the burden of increasing memory demands on AI systems. Shining data down onto processors could lower the energy typically required for data centers, self-driving cars, and even “edge” applications like AI-powered robots, researchers say.
“People are designing all sorts of different AI chips,” says Jae-sun Seo, an associate professor of electrical and computer engineering at Cornell Tech, in New York City. These processors don’t often have room for all the parameters that make up AI models, so the additional data is stored in dynamic RAM (DRAM). The electrical connections commonly used to move the data between the DRAM and the processor create cost and efficiency concerns when systems scale up. “That’s one of the major bottlenecks.”
Optical links move data at high bandwidth with less energy loss than metal wires, but today’s optical receivers undercut that advantage by relying on power-hungry analog circuits to convert light to electronic bits. The group’s new tech would instead receive rapid flashes of digital QR-code-like matrices so that chips can tweak model parameters without those analog circuits, enabling fully digital optical communication that would consume less energy.
“This is a really important problem,” says Dennis Sylvester, an IEEE Fellow who chairs the University of Michigan’s electrical and computer engineering department and was not involved in the work. “It’s got massive commercial implications. This solution is a clever way of dealing with it.”
Jae-sun Seo [left] and Yifan He have developed a receiver that can edit memory in response to QR-code-like arrays of light.Alex Music
How Light “Flips” Memory to Power AI
Processors have a bit of built-in static RAM (SRAM), but not enough to allow an AI model to run independently. While SRAM is the faster of the two memory options, DRAM can store more data in the same footprint.
In the new system, the DRAM sits with the transmitter, and the receiver is part of the processor’s SRAM. The transmitter beams the data to the array of SRAM cells, which in this case are modified to contain photodiodes. Light hitting each photodiode creates a current to flip binary values in the SRAM.
Creating a link between the light and receiver requires calibration, because you can’t expect them to be perfectly aligned or perpendicular to each other. So the chip references a data frame that has information about the expected position of each pixel of data and uses that frame to ensure it can receive the real data, He says. “Ideally the best way is to have direct, point-to-point space between the transmitter and the receiver,” Seo adds, “but even if it’s slightly tilted, we have this calibration circuit.”
For applications in real-world settings, the researchers say they will need to build an optical transmitter that can alter the light matrix millions of times per second, transferring gigabits per second. The transmitter that I saw in He and Seo’s lab is only a proof of concept, emitting a static 14-by-14-bit matrix through a metal mask over the light. The researchers say they are working with optics research groups to build a transmitter that is capable of rapidly changing the matrix.
The Future of Light-Based Memory Links
Michigan’s Sylvester says that the tech in its current form is likely far from commercialization because the individual photosensitive bit cells are larger than SRAM bit cells in conventional chips. Those larger cells mean the chip can fit less memory, a trade-off that he says could cancel out the added efficiency of the light-based approach.
Seo says that it’s part of the group’s ongoing efforts to shrink the bit cells, which can be achieved by optimizing the size of transistors and circuits and leveraging CMOS scaling.
Seo and He are looking at uses for the tech in robotics and other edge applications. One example is in AI-robot-powered warehouses and factories, which could use optical data transmission to save time and energy when updating the AI models in each robot. Additionally, microrobots, which are inherently memory-constrained due to their size, could one day benefit from the tech, though it would require a more size-conscious design.
“Edge AI is a big growth area, and in three, four, five years, you’re going to hear as much about that as you are with data centers, probably, as the intelligence migrates more and more into these devices that we have,” Sylvester says.
Video Friday is your weekly selection of awesome robotics videos, collected by your friends at IEEE Spectrum robotics. We also post a weekly calendar of upcoming robotics events for the next few months. Please send us your events for inclusion.
In just six months, our team turned GENE.01 into a fully functional humanoid platform that can walk, sense and interact. Its full-body multimodal skin perceives touch, proximity, force, and temperature, bringing Physical AI closer to safe and natural collaboration with people. Not a render. Not a concept. This is GENE.01. The future of Physical AI is taking its first steps.
Why create robot intelligence for just one hand, when we could have it learn from many? GEN-1, our latest embodied foundation model, now supports a broad range of end effectors from 5-finger hands, to specialized tools, and everything in between. Each hand is a different sensorimotor interface by which GEN-1 experiences the physical world. Scaling pretraining across thousands of these interfaces teaches GEN-1 a universal physical common sense that transfers to new hands and new ways to grasp, push, pull, twist, and more.
And to illustrate this concept, a surprise spatula.
This paper presents the design, fabrication, and flight validation of a flat-packable flying wing built primarily from corrugated cardboard. The aircraft is manufactured from three laser-cut sheets and assembled through a fold-and-lock architecture that forms load-bearing wing structures with minimal tooling and no permanent fasteners. The full airframe can be assembled in under 15 minutes, demonstrating strong potential for rapid deployment, low-cost logistics, and scalable field use.
Together with Niantic Spatial and Nvidia, [we] can now scan a real deployment site with off-the-shelf hardware, reconstruct it into a photorealistic Gaussian splat, and run massively parallel RL [reinforcement learning] training. The policies trained in our Gym environment then transfer zero-shot to the real robot and environments they were trained for. This enables faster deployment of more capable and robust policies for the end user.
I think the right thing to do here is only post the disclaimer included with this video: “This film is a conceptual creative production, and certain scenes are presented for demonstration purposes only and do not represent the actual in-store operating process. The final store environment, robot appearance, and functionality are subject to the actual deployment. During actual operations, the robot will autonomously perform only designated preparation steps for specified ice cream products, and its hands will be fitted with protective gloves that comply with applicable food safety requirements.”
Drone delivery is now an essential part of the South West London Pathology (SWLP) modernization agenda. Since February 2026, our highly automated aircraft have been delivering urgent NHS samples across south west London, with service up to 85% faster than ground transport. We are thrilled to be part of this initiative, supporting clinicians in providing timely, effective care for patients and contributing to a greener, more resilient NHS.
Take a closer look at what’s next for the Aurora Driver. Designed to move freight farther, faster, and more efficiently, this next generation of the Aurora Driver delivers greater performance, built to last one million miles and cut hardware cost in half.
How does a robot learn to recognize an object it’s never encountered? In this case, a demo can be worth a thousand words. Short human demonstrations can be used to create fully automated training datasets, sidestepping the prompting limitations that hold back vision-language models. Rather than describing objects with language, the system tracks what a person touches and manipulates during a demo, follows those objects through time, and clusters detections to handle objects merging or splitting apart in the scene. This bypasses a core weakness of VLMs, which struggle to reliably detect unusual or novel objects even with repeated, carefully engineered prompts.
Like many engineers, Sarah Downs says she knew she wanted to pursue a STEM career from a young age. As a teenager, she discovered robotics through her Tulsa, Okla., middle school’s First Lego League team, and she fell in love with the field, she says. Downs participated in the international robotics program from 2014 to 2016.
Watching PBS specials on NASA Mars rovers Spiritand Opportunity, and seeing the live broadcast of the Curiosity rover launch in 2011, inspired the teen to dream of a career working with NASA.
Sarah Downs
MEMBER GRADE
Graduate student member
UNIVERSITY
Texas A&M University in College Station
MAJOR
Electric engineering
This year the IEEE graduate student member achieved that dream. For her final project as a master’s degree candidate in electrical engineering at the University of Tulsa, she worked on an algorithm in collaboration with NASA and the U.S. Air Force.
The algorithm she developed enables a robot assembling satellites in space to insert an antenna into the correct spot, addressing robotics’s classic peg-in-hole problem of inserting an object into its corresponding hole.
Now a Ph.D. student in electrical engineering at Texas A&M University in College Station, Downs is continuing her research on satellite assembly and manipulation “but on a much larger scale,” she says.
Following a childhood passion
Downs grew up in the Tulsa area. Her father, who died from a heart attack in 2015 when she was 13, was a safety advisor in the oil and gas industry. Her mother stayed home to take care of her brother, who has autism. After her father died, her mother went back to college to earn a bachelor’s degree in business so she could support the family.
“We didn’t have much income, and my mom was always worried about money,” Downs says. “That made me more aware of having a successful career, in a monetary sense.”
From then on, whenever she considered her future career, having a decent salary to support the family was high on her list.
By pursuing a career in robotics, she says, she can follow her passion while obtaining financial security.
In high school, Downs joined the First robotics club, where she found herself drawn to the electrical components used in the machines she and her classmates built.
During her final two years of high school, she participated in an extension program at Tulsa Tech, a training school. She spent half her day in high school classes and the other half taking engineering courses at the vocational school.
After graduating in 2020, she accepted scholarships to attend the University of Tulsa. She began her freshman year at UTulsa not knowing whether she wanted to major in electrical or mechanical engineering, she says, adding that her love of working with small systems helped her choose EE.
For her senior year capstone project, she and two of her classmates designed a lunar lander exhibit for the Tulsa Air and Space Museum. They created an interactive game that simulates missions on lunar and martian surfaces. Four celestial bodies—the moon, Venus, Mars, and Titan—are listed across three computer monitors. Using a game controller, museum visitors can explore the virtual surface of each one. The exhibit is still on display.
Downs earned her bachelor’s degree in electrical engineering in 2024 and continued her education at the university’s EE master’s degree program.
Both more and less complicated than people think
When Downs began her graduate studies, she was supposed to be part of a NASA robotics project for two years. But when a delay in government funding postponed the project’s start, she instead spent her first year in the school’s Institute for Robotics and Autonomy, then newly launched. Its main focus is developing robots to assist people who have mobility challenges.
Inspired by her grandmother, who was wheelchair-bound due to severe arthritis, Downs developed a robotic arm that helps older people and wheelchair users live independently. The arm was able to identify and place objects in the appropriate locations inside the home, such as unloading certain groceries from a shopping bag and placing them on a shelf or in separate containers.
Before the start of her sophomore year in 2025, the NASA project finally secured government funding. She developed a robot that achieves the peg-in-hole task without using any vision systems. Typically, cameras help guide robots’ satellite-assembly work. But in the harsh, remote environment of outer space, cameras might malfunction or encounter delays.
“Don’t stop asking questions. Especially in engineering, don’t pretend like you know everything, because science is about constantly wanting to learn and listen.”
Rather than using cameras, Downs’s robotic arm deploys a force-based insertion process to sense position and orientation of objects in the arm’s environment. The robot loosely grips an antenna and, with a torque sensor on its gripper, “feels” the force feedback of where the satellite and antenna are in relation to each other. The robot then guides the antenna assembly into a target opening on its satellite and maintains the position during adhesion.
Adding to the complexity, the robot performs its task in zero gravity.
“Without gravity, you now have to consider the arm’s reaction torques on the satellite to avoid flinging it into space,” Downs says. Any motion from the arm during the insertion process, especially from increased forces, could cause the satellite to continue movement in that direction.
To combat that, Downs is performing calculations for the project to direct targeted reverse thrusts and counter the force of the robot’s motions.
Her graduate project captures the simple yet complex nature of robotics that she finds fascinating, she says.
“I think robots are both more and also less complicated than people think,” she says. “Really, all you need to start programming a robot is its Denavit-Hartenberg parameters, and you can do a lot with that,” she says, referencing the four values used to describe the position and orientation of a robotic arm and manipulators. Even with different grippers and degrees of freedom, “fundamentally, all robot manipulators start there,” she says.
“But,” she adds, “we’re still learning so much about how robots interact with their environment. Even something simple to us, like manipulating a pen, is still incredibly complex for robots.”
Her thesis advisor is Robert Ambrose, a NASA veteran who launched the RAD Lab in 2022. The IEEE member is set to serve as associate director of the school’s Space Institute, due to open this year in Houston. The research facility is being built next to the Johnson Space Center.
After earning her Ph.D., Downs says, she hopes to one day work for NASA, developing rovers that collect samples from Mars or robotic arms that perform tasks on space stations.
To learn more about robots, check out IEEE Spectrum’s guide.
Getting out of the engineering bubble
Downs joined IEEE in 2020 as a freshman at UTulsa to get more involved in electrical engineering events on campus. At the time, the COVID-19 pandemic kept clubs and organizations from meeting in person.
She was active in her school’s IEEE student branch and was elected as its 2022–2024 president. Under her leadership, the branch went from having a few events to hosting one every two weeks.
They included lunch-and-learn sessions and dinners that connected students with professional engineers and the university’s alumni. Downs also organized hands-on workshops on soldering, 3D printing, CAD modeling, and résumé-building.
Her efforts helped increase the branch’s executive board membership from roughly five students to 25 in 2023. The same year, her soldering workshop attracted about 80 students.
She says she enjoyed working with IEEE, especially “engaging with alumni and learning from engineers.”
IEEE is a great resource for networking opportunities, she says, noting that “during the COVID-19 pandemic, engineering students stayed in their bubbles.” IEEE events helped the students make connections that could serve them well, she says.
“Networking is very important, especially in today’s tough job market,” she says. “It’s a lot about who you know and how people observe your work ethic.”
Downs, who now serves as an IEEE graduate advisor for UTulsa’s student branch, says she has seen firsthand how the school’s student branch network has benefited its student members.
“A lot of them have found jobs” because of IEEE, she says.
The working and networking of an engineer
As the IEEE graduate advisor for UTulsa’s student branch, Downs noticed that many engineering undergraduates finish college without any hands-on experience, whether it be a project or an internship.
“Their résumés are very sparse, and they have no proof of their technical skills,” she says. She herself completed a facilities engineering internship at Tulsa International Airport’s American Airlines maintenance facility after her sophomore year of college. And she was an electrical engineering intern at Flight Safety International outside Tulsa after her junior year and after she graduated. The company designs, builds, and maintains its own flight simulators.
Her advice to undergraduates is to hone and demonstrate both their hard and soft skills by working on research projects or even personal passion projects.
“A Raspberry Pi doesn’t cost that much, and you can start working with that immediately,” she says. Students also can take part in engineering interest groups and professional organizations at their school, she adds.
“Put yourself out there and join a research team,” she says. “It’s a great way to show people that you’re a good person to work with and you’d do a good job in the field.”
She adds that it’s also a fine way to keep learning—which is what drew her to a field that has developed only within the past century.
“We’re still constantly learning about robots,” she says.
“Don’t stop asking questions,” she advises students. “Especially in engineering, don’t pretend like you know everything, because science is about constantly wanting to learn and listen.”
Video Friday is your weekly selection of awesome robotics videos, collected by your friends at IEEE Spectrum robotics. We also post a weekly calendar of upcoming robotics events for the next few months. Please send us your events for inclusion.
In this work, we present a systematic evaluation of contemporary humanoid technology for laparoscopic surgical tasks. We develop a humanoid-based laparoscopic teleoperation framework using general-purpose instruments and assess its capabilities through benchtop characterization, dry-lab user studies spanning diverse surgical experience levels, and in vivo porcine studies. Across these evaluations, we quantify technical feasibility, task performance, and clinical readiness relative to established surgical platforms. Together, our study provides an evidence-based assessment of the current capabilities and limitations of humanoids for surgical applications, highlighting both their promise and the key technical challenges that must be addressed before clinical deployment.
Today, we preview ACT-2, the first robotics model to achieve reliability by unifying broad generalization with high performance.
Sunday also has this 3-hour video (!) of Memo folding laundry in “never seen environments.” Let’s just not ask, because we almost certainly don’t want to know.
Spot is not the first quadruped to try its legs at last few meters package delivery, but the challenge is not really those last few meters—it’s going to be not driving the human coworker nuts, is my guess.
Quadrupedal locomotion in complex environments requires multiple motor skills, stable gait transitions, and perceptive control over a broad range of speeds. APT-RL (Action Pretrained Transformer-based Reinforcement Learning) is a unified framework for high-speed, multiskill locomotion. A single policy selects and transitions between gaits and motor skills using only onboard perception and computation. In real-world experiments, KAIST HOUND traversed stairs, hurdles, stepping-stones, gaps, and fallen branches. It reached an instantaneous peak speed of 4.25 meters per second while traversing a 60-centimeter step and 6 m/s during a drop-down transition on a three-step staircase.
Following Atlas’s first-of-its-kind live performance at the FIFA World Cup 2026, we caught up with Seth Davis, senior program manager, to learn how this demonstration came together and what it takes to succeed in the field (and on the pitch).
No teleoperation. No cuts. Long take. One of the world’s few complete demonstrations of long-horizon mobile manipulation, bringing fully autonomous humanoid robots another step closer to us.
We saw some footage of this last week, but here’s a much better video.
Wing-propelled diving birds flap their wings to move through air and water, yet the wing morphology and kinematics that enable this behavior remain poorly understood because of the difficulty of collecting in situ data. The impact of flapping frequency, wing size, and stiffness on locomotion in—and transition between—the two media are still unknown. We compared data from diving birds against experiments using a flapping-wing robot capable of flying, swimming, plunge diving, and exiting the water. We show that frequency adaptation, flexible wings, and powerful actuation enable seamless transitions without folding wings or legs, that large wings enhance flight without substantially reducing underwater efficiency, and that tail-body distance and egress angle affect water exit. These results clarify how birds (and robots) balance multifluid locomotion constraints.
There are many words that I would never, ever use to describe a drone. Stealthy. Subtle. Whatever the opposite of obnoxious is. Much of this is because of the giant angry bee sound that drones tend to make, but it’s also the way that they look in flight: With uncannily linear movements and an even less canny ability to hover perfectly still, they tend to draw the eye as affronts to nature.
In a paper presented this week at Robotics Science and Systems 2026 in Sydney, roboticists from Northwestern University, Evanston, Ill., demonstrated a drone called Phantom Twist that is essentially invisible to humans, being an order of magnitude more difficult to see in flight than a typical quadrotor. They accomplished this with the aid of computational design, and while the resulting hardware is, I would argue, also an order of magnitude more of an affront to nature than a typical quadrotor represents, it’s pretty amazing how well it works.
Phantom Twist spins so fast, it’s practically invisible.Michael Rubenstein/Northwestern University
The trick here is easy to see, even if the drone isn’t. By spinning in flight at between 15 and 25 hertz, Phantom Twist takes advantage of humans’ decidedly mediocre visual system to turn a solid spinning object into an opaque smear. Human eyes take some amount of time (typically about 100 milliseconds) to integrate what we see before sending the full scene off to our brains for processing. Moving objects can cause problems for this system, because if the movement is fast enough, our eyes are forced to average that motion across the scene, combining it with whatever is in the background and resulting in a transparent blur. This effect is called persistence of vision. For something that spins like Phantom Twist, that motion blur comes from the drone’s rapid rotation, and it works because most of the drone is cleverly designed to be empty space.
Drones that spin in flight are nothing new—we’ve covered a bunch of them in the past, including Picolissimo and any number of samaradrones inspired by the spinning flight of maple seeds. What makes Phantom Twist unique, and also very odd, is that the design was computationally optimized for low visibility.
Controlling how drones like this fly
Before we get into that, though, a quick note about how drones like this can even fly controllably, because it’s not at all obvious. With just a single motor and no control surfaces, the only possible control input is through the motor itself, and by pulsing the motor speed up or down at just the right time during each rotation, the drone can translate in any direction. Altitude control comes from changing overall motor thrust, and the drone‘s spinning nature makes it passively stable.
Carbon fiber rods connect batteries, a controller, some counterweights, and a motor and propeller. The research robot also includes optical tracking tags.Michael Rubenstein/Northwestern University
The bits that you need for this kind of drone include the motor and propeller, a couple of batteries, a controller, some counterweights (which could be replaced with more batteries or payload), 0.8-mm carbon fiber rods to tie it all together, and a connector for the handheld launcher that gets the whole thing up to speed. The actual arrangement of these components is surprisingly flexible, and that’s where the invisibility comes in.
“The design space is high dimensional,” explains Northwestern’s Michael Rubenstein. “It’s very difficult for a human to reason through all the trade-offs between the physical constraints required for stable flight and the visual appearance of the spinning drone, and I don’t think we would have easily arrived at this low-visibility design ourselves.”
The visibility (or not) of Phantom Twist is primarily driven by the extent to which different components line up with each other from the perspective of someone looking at the drone. The more components that line up with each other as the drone flies, the less background you see through the spinning drone, and the more visible the drone becomes. Because you might be looking at the drone from a number of different angles, and also because the drone has to be stable enough for controlled flight, there are a bunch of different things that need to be optimized all at once, which is why computational design is effective here.
Phantom Twist’s final design was generated using an iterative optimizer which had a goal of minimizing a metric called learned perceptual image patch similarity, or LPIPS, while making sure that the design could still physically work. LPIPS is the difference between two images: a background image, and a background image with an overlay of the simulated spinning drone. The smaller that difference is, the more invisible that design is. It’s tricky for a human to consider all of the variables at once, but Rubenstein says that the final design does make intuitive sense, because “the automated pipeline prefers placements where components don’t visually overlap as it spins, or where the components are too close to the center of rotation.”
Two iterations of Phantom Twist drones are shown with their handheld launching mechanisms. The better-optimized version [bottom row] relocates the launcher interface to remove components that are too close to the central axis, making them more visible.Michael Rubenstein/Northwestern University
Out of a starting set of around 20,000 feasible Phantom Twist configurations, the optimized design (the one that you see or don’t see in the pictures and videos) has a LPIPS score of 0.0104. A human-designed Phantom Twist is about twice as visible, with a LPIPS score of around 0.2, and a conventional quadrotor (of the same size) would be over 10 times more visible. And there’s still a bit more optimization that could be done with the electrical wiring as well as increasing the baseline transparency of the components themselves.
Phantom Twist is currently controlled using an optical tracking system, which means that it’s not yet capable of flying outside of a controlled environment. But Rubenstein has built other drones along similar principles in the past, which have successfully flown outside, and he’s optimistic about using those techniques to break Phantom Twist out of the lab. The spinning behavior might even enable some useful sensing capabilities, he says. “An interesting possibility is mounting a camera on the spinning body. As the vehicle rotates, it could capture imagery in every direction, effectively creating a 360-degree view of its surroundings that could be used for onboard navigation and control.”
As for what a drone like Phantom Twist could be used for—assuming that the sound can be mitigated somewhat (and there are potential approaches to making that happen), a stealthy microdrone could do all sorts of things with covert surveillance being the most obvious application. For his part, Rubenstein says that he’s personally excited about the potential for watching wildlife, “where a less-intrusive drone could observe animals while minimizing its impact on their natural behavior.” The elephants in particular would certainly appreciate that.
Large language models gave artificial intelligence a working recipe. Pretrain a large model on broad data, and general capability follows. Robotics has no such recipe. Robotics systems have long been assembled from separate perception, planning, and control parts that rarely add up to intelligence a robot can carry from one task to another, or one machine to another. The central problem in embodied AI is to find the equivalent recipe, and the field does not yet agree on what it is.
X Square Robot, a Chinese embodied-AI company, has made an unusually explicit bet. It argues that the recipe is an integrated stack, spanning the data a robot learns from, a world model for predicting changes in the physical world, and an action model that brings together perception, planning, reasoning, and decision-making to generate executable robot behavior. The company also believes that the stack should be built and released in the open.
X Square Robot shares its vision of bringing robots into real homes.X Square Robot
X Square Robot’s embodied AI stack
What holds the stack together is a small set of principles rather than a single overarching model.
The first is that the basic unit of robot data is an interaction, not a trajectory; a demonstration is successful only if it changes the world as intended, not simply because the joints moved.
The second is that pretraining should yield usable capability, not just an initialization for later fine-tuning.
The third is that behavior should be modeled around physical events rather than fixed slices of time.
These principles make the layers interdependent, since the same robot-free data that trains the action model is also structured to feed the world model. It is worth being precise, though. The company describes the world model and the action model as complementary but independent model families that share a code base. Both sit within its broader World Unified Model, which it has presented as an architecture for training vision, language, action, and physical prediction together.
Robot learning data: Engineering for quality and cost, not scale
For the X Square Robot team, one of the biggest constraints on general-purpose robots is the cost and quality of interaction data, not the number of parameters. To address that, the company built its Universal Manipulation Interface (UMI) data collection system, QUANXTA Zero Series. It works by collecting demonstrations from people wearing a rig with dual grippers rather than teleoperating a robot. This approach is not itself new, and builds on established methods for robot-free data capture. What sets it apart are two engineering choices.
X Square Robot emphasizes data quality control, recording trajectories and replaying them on a real robot, with only those that actually complete the task counted as valid.X Square Robot
The first is quality control, and it is the most distinctive part. Rather than accepting recorded trajectories as they are, the system runs a closed inspection loop, and its notable step is physical playback. A sample of trajectories is replayed on the real robot, and only those that actually complete the task count as valid. That makes the validity rate a measured quantity rather than an assumption. For example, a gripper that closes a fraction of a second too early still looks like a grasp in the data, yet it has pushed the object away, so it shouldn’t be classified as valid. A smaller clean dataset can be worth more than a larger noisy one.
The second choice is how lower-cost human data and scarce robot data are combined. The company pretrains on a large volume of robot-free demonstrations to build general representations, then adds a small amount of real-robot data as an anchor to the specific machine’s dynamics. It reports that this reaches performance comparable to an all-robot dataset at roughly a 20-fold lower cost of collection, driven mainly by how much cheaper the wearable rig is than a teleoperation setup.
The resulting dataset is deliberately model-agnostic, formatted to feed both action models and world models. The caveat is that the strongest results are measured on the company’s own robots and data-collection pipelines. Broader independent testing will help confirm and extend these promising results across a wider range of settings.
A world model organized around events
In developing its world model, called WALL-WM, X Square Robot took a differentiated approach. Most action models predict a fixed-length chunk of motion from the current image and instruction. That is convenient, but it segments behavior into fixed-duration windows, so the boundaries fall where elapsed time dictates rather than where one action ends and the next begins. WALL-WM instead treats an action-grounded semantic event as its unit: a coherent piece of behavior such as reaching, grasping, or placing, something that can be named in language, seen in video, and executed as motion.
X Square Robot’s world model, called WALL-WM, treats an action-grounded semantic event as its unit: a coherent piece of behavior such as reaching, grasping, or placing, something that can be named in language, seen in video, and executed as motion.X Square Robot
WALL-WM’s design reflects a specific concern about not discarding what large video models already know. To achieve that, a text-to-video model is coupled to a freshly initialized action network that reads from the video features without overwriting them, which preserves the visual prior. From that one process, it offers two modes. An event mode runs in variable-length segments and suits reasoning over long horizons, while a fixed-length mode produces the steady, real-time output a controller needs. That places WALL-WM between mainstream chunk-based action models and pure video world models, keeping the predictive character of a world model while still yielding executable control.
In a series of experiments, the company relied on a generalization test that is more specific than most. A model trained on a limited dataset was evaluated on long-horizon tasks in unseen settings and, on the company’s real-robot benchmark, reportedly outscored baselines that had been fine-tuned on related data. That is a meaningful result if it holds. For now, it is measured on the company’s own benchmark. With the code now being released, the broader community will have the opportunity to test, reproduce, and build on them across more settings.
A policy that runs before fine-tuning, and action tokens with meaning
The action layer carries two connected ideas. The first is a requirement the company sets for itself with Wall-OSS-0.5, its vision-language-action model: The pretrained model should run on a real robot before any task-specific fine-tuning.
The interest is less in the scores than in the design behind them. The model trains three objectives together, namely discrete action tokens, language grounding, and continuous action generation. And it keeps gradients flowing through all of them rather than freezing parts of the network as some rival designs do. It’s also a more strict method, since it reports untuned behavior such as approaching, grasping, and recovering, including on a deformable task held out of training.
As part of X Square Robot’s Wall-OSS-0.5 vision-language-action model design, the pretrained model should run on a real robot before any task-specific fine-tuning. X Square Robot
The second idea is the action interface itself, called X-Tokenizer. Most systems that turn continuous motion into discrete tokens produce codes that the language model cannot interpret. X-Tokenizer reframes tokenization as learning a semantic interface, so that the top-level code stands for the intent of a motion while lower-level codes carry finer detail, all aligned with the language model’s own features.
A useful consequence is stability. Adding noise to an action barely moves the intent code, which is what lets one tokenizer to be reused across robots without re-tuning. The tokenizer inside the production action model is a related variant of this approach. Together, the two ideas give the action layer something rather powerful: capability that transfers.
The future of embodied AI stacks
X Square Robot is betting that its unique approach combining three layers, each specialized in solving a key part of the problem, will stand out from other embodied AI stacks. The physical-playback step that grounds data quality is uncommon and sensible. The reframing of world modeling around events, with one backbone serving both reasoning and control, is a genuinely distinct approach. And the pairing of a deployable pretraining standard with a tokenizer designed as a semantic interface gives the action layer unusual coherence.
X Square Robot’s valuation has climbed above 20 billion yuan (about US $2.9 billion), suggesting that investors increasingly view data infrastructure, foundation models, and scalable training systems as long-term differentiators in embodied AI.
The next phase will bring broader validation. Much of the current evidence comes from X Square’s own robots and benchmarks. With the world model code now being made public, and as the community begins to test, reproduce, and build on the work, the reported capabilities will be tested across more robots, tasks, and settings.
X Square Robot’s recent funding rounds reflect similar confidence. The company’s valuation has climbed above 20 billion yuan (about US $2.9 billion), suggesting that investors increasingly view data infrastructure, foundation models, and scalable training systems as long-term differentiators in embodied AI.
What’s next for X Square Robot
To learn more about its future plans, the following Q&A with the X Square Robot team further explores the company’s technology, strategy, and vision.
What made now the right moment, technically, to commit to this stack? What recently became possible that wasn’t possible a couple of years ago?
It is not one breakthrough but several trends maturing together. Foundation models gave us a shared representation across vision, language, and action, so we can model what a robot sees, what it is asked to do, and how its actions change the world in one framework, rather than as separate perception, planning, and control modules.
Compute and infrastructure are finally sufficient for large-scale pretraining over long-horizon, multi-embodiment data. Just as importantly, we realized that data, not model size, is the real bottleneck for general robots—what is scarce is diverse, high-quality, reproducible interaction data. And world modeling has become practical. The useful question is no longer how to predict a few seconds of video, but how to understand the ways actions change objects, contacts, and task states. Two years ago these ingredients existed separately. Today they are mature enough to work as one system.
“We realized that data, not model size, is the real bottleneck for general robots—what is scarce is diverse, high-quality, reproducible interaction data. And world modeling has become practical.”
Your data system captures demonstrations with a wearable VR rig and custom grippers rather than teleoperating robots. What was wrong with standard teleoperation?
Teleoperation is built around controlling the robot. It forces the operator to work within the machine’s kinematics, latency, and viewpoint, and the resulting demonstrations are slower, stiffer, and less diverse. We built our system around capturing human skill instead. Manipulation is really about contact, timing, finger coordination, and recovery, not just the path the hand takes, and a wearable rig records those before the behavior is compressed onto one particular robot. It also breaks teleoperation’s expensive scaling law, in which every demonstration needs a robot.
People can generate rich data independently of any robot, and the crucial property is that those demonstrations can still be replayed and executed on a physical robot through the model. Mobility is convenient, but that replay is the real point, because it is what lets the same data be reused across different platforms.
In X Square Robot’s approach, demonstrations can be replayed and executed on a physical robot through the AI model, allowing the same data to be reused across different platforms.X Square Robot
X Square Robot reports that its pipeline has roughly an 85 percent data-validity rate. Why is quality control such an underrated bottleneck?
Because errors in robot data are far more expensive than in language data. A small timing or contact error can change what a demonstration means. If a gripper closes a fraction of a second too early, the motion still looks like a grasp, but physically it has pushed the object away. A dataset that mixes failures and accidental successes teaches ambiguity, not skill, because the real unit is the interaction, not the trajectory.
So we run automated inspection, kinematic checks, and physical replay, where we play a sample of trajectories back on the real robot and count only the ones that actually complete the task. Data quality sets the ceiling on how good a policy can be. In our experience a smaller, cleaner dataset often beats a much larger, noisier one, which is why we treat quality control as part of the model, not a preprocessing afterthought.
The model runs in both “event mode” and “chunk mode.” When does each matter?
Both matter, for different reasons. The physical world changes through events—when contact occurs, a grasp forms, or an object slips—not in fixed-frame windows. Event mode concentrates the model’s attention on those moments, and it matters most for long-horizon tasks, like clearing a table, where progress is a sequence of semantic events rather than a smooth stream. It runs in variable-length segments that follow the task rather than a clock. Chunk mode matters for deployment. Real controllers need a stable, real-time interface, and fixed-length chunks integrate cleanly with existing control systems.
We organize learning around events in the first place because a fixed window can split one motion in half or merge two together, which turns training into short-horizon pattern matching and weakens the model on long tasks. So the world model’s job is to connect event-level understanding, which is where the reasoning happens, with a fixed-length output a real robot can actually run.
Why make “deployable before fine-tuning” the criterion?
Pretraining should produce capability, not just a good starting point. If a model is only useful after heavy fine-tuning, then most of the intelligence still lives in the downstream supervision, not in the foundation model. Deployable before fine-tuning is a more honest test of what pretraining actually learned. A well-pretrained robot should already know how to approach, grasp, move, avoid obstacles, and correct itself. Fine-tuning should adapt it to a specific task or robot, not create the ability from nothing. It is also a practical requirement. A robot in a home or a workplace shouldn’t need a brand-new dataset and a new policy every time the task changes, so a foundation model that already carries general skill, and some ability to recover, is the minimum bar for something genuinely useful in the real world.
What is the most challenging part of cross-embodiment learning?
Robots differ in control frequency, delay, compliance, sensing precision, and contact dynamics, so the same instruction can require different action decompositions and recovery strategies, and a behavior that works on one arm cannot simply be copied to another. Cross-embodiment learning needs an intermediate abstraction, lower than language but higher than joint angles: how you approach an object, how you make contact, how you apply force, and how you recover from a mistake.
When we say cross-embodiment, the main capability we mean is multi-embodiment generalization: transferring across robots, training on many embodiments at once, and adapting to different kinematics. Human-to-robot transfer and other techniques are specific approaches to that goal.
“A robot in a home or workplace shouldn’t need a new dataset and policy every time the task changes. A useful foundation model should already carry general skills and the ability to recover.”
What would you most like to see other researchers attempt to reproduce or stress-test?
Three things, above all. Whether event-level representations really generalize beyond our own datasets, across more tasks, scenes, objects, embodiments, and failure conditions. Whether pretraining stays effective on robots the model never saw during training, or whether its capability is still too tightly coupled to what it has already seen. And whether real-robot evaluation can become a shared language for the field, so that we compare not just success rates but the reasons systems fail, where an instruction was misread, where perception broke down, or where recovery fell short. Robotics has been driven too often by impressive demonstrations, and real progress comes from results that are reproducible and diagnosable.
What capability is still missing before robots become dependable in homes?
Benchmarks measure competence, like whether a model can finish a task. Homes demand reliability, safe and consistent operation over time in a place that changes every day, with objects moving, instructions that are vague, and people interrupting. The missing piece is not a higher one-time success rate: it is robust recovery. A dependable home robot has to know when it is uncertain, when to slow down, when to ask for help, and how to bring the world back to a safe state after it drops something or misunderstands a request.
In a real home, failure recovery matters more than raw success, because the home does not reset itself. Homes also demand careful personalization, learning a household’s routines and preferences over time, with safety and trust as first principles. That combination, not any single skill, separates a capable demonstration from a robot people can live with.
X Square Robot’s approach is that, in a real home, failure recovery matters more than raw success, because the home does not reset itself and it demands careful personalization, with safety and trust as first principles. X Square Robot
How do the open-source components fit into X Square Robot’s World Unified Model direction?
We see these releases as layers of the World Unified Model direction rather than isolated projects. Wall-OSS-0.5, the action model, asks whether an open vision-language-action model can gain directly measurable capability from large-scale pretraining, so it is the capability layer. WALL-WM, the world model, asks how a robot should understand change in the world, shifting from fixed windows to event-level modeling, so it is the representation layer. The data system supplies the interaction data that both of them learn from.
Together they form a loop in which models produce capability, world models organize understanding, and the open-source community drives reproduction and improvement. World Unified Model is the broader architecture those layers support, bringing vision, language, action, and physical prediction together.
We are releasing these pieces openly because embodied intelligence cannot be solved by one organization; it needs many embodiments, many real tasks, and broad feedback, and the long-term goal is a stack that keeps learning and ultimately moves robots from laboratory demonstrations toward reliable everyday use.
Video Friday is your weekly selection of awesome robotics videos, collected by your friends at IEEE Spectrum robotics. We also post a weekly calendar of upcoming robotics events for the next few months. Please send us your events for inclusion.
For the first time, two full teams of humanoid robots played an 11-vs-11 soccer match on hardware, bringing one of robotics’ most ambitious long-term visions closer to reality. Never before have two full-sized humanoid robot teams played a soccer game against each other.
Engineers at MIT and EPFL in Lausanne, Switzerland, have designed a robot that can swim underwater and flap out of the water to continue flying through the air, much like a diving bird. The robot can help scientists study the mechanics that enable these actions in aquatic aviators and may help launch a new class of aerial-aquatic drones and vehicles.
We’re excited to announce our breakthrough robotic hands for the NEO platform: hands that match or exceed human-level dexterity, strength, safety, and reliability. Designed from the ground up, these 25-DoF hands combine 25 fully actuated degrees of freedom with a tendon-driven system, rich tactile sensing, and built-in compliance. The result is a hand capable of true in-hand manipulation, precision tool use, and delicate interaction.
Atlas arrived pitchside at NYNJ Stadium in front of 80,000 people gathered to see Brazil vs. Norway. After performing some of the sport’s most memorable player celebrations, Atlas helped kick off the second half by delivering the match ball!
Navigating discrete terrain such as stepping stones remains a major challenge for legged robots. Conventional approaches often rely on dense environment reconstruction from cameras or lidar, which can be affected by latency, occlusions, and significant computational overhead. We show that proximity sensors integrated into the bottom of a quadruped’s feet enable safe, terrain-seeking autonomous locomotion.
On this holiday, Digit is on grill duty. It turns out precise force control is good for more than payload handling. Happy 4th of July from all of us at Agility.
We’ve created GEN-1, our latest milestone in scaling robot learning. We believe it to be the first general-purpose AI model that crosses a new performance threshold: mastery of simple physical tasks. It improves average success rates to 99 percent on tasks where previous models achieve 64 percent, completes tasks roughly 3x faster than state-of-the-art, and requires only one hour of robot data for each of these results. GEN-1 unlocks commercial viability across a broad range of applications—and while it cannot solve all tasks today, it is a significant step toward our mission of creating generalist intelligence for the physical world.
Reachy Mini is becoming your real AI companion. The Conversation App makes it able to talk fluently with you, help you with your to-do list, remind you of important tasks, and even chat about music. Long-term memory, voice interaction, always ready to help.
This is very delicate work, although the real challenge would be picking those nuts out of a jumbled bin full of randomly sized nuts, which is how most of us live our lives.
In this presentation, Brendan Schulman, vice president of policy at Boston Dynamics, outlines the critical role of government engagement in driving the success of the humanoid robotics industry. He demonstrates how legged robots like the Spot quadruped and Atlas humanoid are moving beyond factory settings to deliver real-world value in infrastructure inspection, industrial manufacturing, and public safety. Schulman highlights the intersection of AI and robotics, showcasing how large behavioral models and reinforcement learning enable robots to navigate slippery floors and autonomously avoid workplace hazards. Ultimately, he calls for a proactive national robotics strategy focused on workforce training, safety standards, and ethical frameworks to support supply chain resilience and global competitiveness.
Borys Drozhak has a vision: a front line almost free of humans, patrolled by flying drones and ground robots, and continuously monitored by AI-controlled sensor networks. And it’s not a pipe dream. Ukrainian roboticists have made major strides in that direction over the past four years. Remotely controlled ground vehicles fitted with machine guns and grenade launchers now patrol the no-man’s-land straddling the front, part of a robotic legion that has stymied Russia’s territorial ambitions so far this year.
Drozhak is a co-founder and CEO of RoverTech, which manufactures the Zmiy, one of Ukraine’s most successful ground robots. Zmiy, Ukrainian for snake, is an 800-kilogram (1,700-pound) rover, 2.15 by 1.5 meters in size, with 75-centimeter diameter wheels. The Zmiy comes in various configurations—for demining, logistics, fighting fires, firing a machine gun, or launching grenades.
According to Drozhak, the uncrewed ground vehicle (UGV) is a record-breaker among Ukrainian ground robots. It’s engineered to be nearly noiseless and emit as little heat as possible, helping it to elude Russia’s intelligence, surveillance, and reconnaissance (ISR) drones. As a result, a Zmiy rover completes on average 57 missions across the kill zone before being destroyed. The kill zone is the roughly 35-kilometer-wide swath of land that straddles the front line; its width is variable and determined mainly by the growing range of the drones.
“Usually, a UGV on the battlefield lasts about seven missions,” Drozhak says. “The Zmiy is quite a bit bigger and stronger” in comparison with most other UGVs, “and can make it back even if two of its wheels get destroyed.”
Drozhak is a software engineer turned roboticist whose story is echoed everywhere in the Ukrainian defense establishment. Before the Russian invasion, he was living a quiet life in Ireland, working for an international software development firm. He returned home shortly after the war began to help defend his homeland. Together with his friend, Vasyl Korenovskyi, who had been a mining engineer, he founded RoverTech with the goal of building robots to perform some of the most dangerous tasks in the war zone. In 2023, they rolled out their first product—the Zmiy de-miner. Earlier this year, one of RoverTech’s assault UGVs was part of a widely reported operation that forced a group of Russian soldiers to surrender without the presence of any Ukrainian troops. Such feats, Drozhak insists, are not rare on Ukrainian battlefields these days.
UGVs are the latest chapter in the military-technology race spurred by the war in Ukraine. Scores of Ukrainian startups have developed dozens of different small ground robots, each with typically multiple variants, over the past three years. They’re mostly replacing human-driven tanks and other military vehicles that used to crisscross the war zone. These remotely controlled robotic vehicles cost a few tens of thousands of dollars apiece compared to millions for a traditional tank, and they can be tweaked and modified in frontline workshops to serve the most urgent needs.
Zelenskyy Orders Up 50,000 More UGVs
In April, Ukraine’s President Volodymyr Zelenskyy signed an order for the government to procure 50,000 UGVs for Ukraine’s military forces by the end of 2026. That’s more than three times as many as the government purchased in 2025 and a massive increase from the 2,000 procured in 2024, according to defense analyst Marc C. Lange.
The rise of UGVs, Lange explains, is a direct response to the warfighting revolution ushered in by the speedy evolution of uncrewed aerial vehicles that came to define the war in Ukraine.
As the number of drones zooming above the front line rose and their range increased, the battlefield became completely transparent. Today, anything that enters the kill zone gets hit by a first-person view (FPV) kamikaze drone within minutes.
“Any armored formation, any resupply and logistics vehicle, and any manned formation anywhere near the edge of the battle area has between seconds to a low amount of minutes before it gets turned to dust,” Lange says. “The Ukrainians were losing drivers. Traditional methods of evacuating injured soldiers became impossible. That space is basically unsurvivable.”
Ukraine, suffering from a shortage of infantry, has taken that problem more seriously than Russia, which has a larger pool of fresh recruits to draw from. UGVs began ferrying supplies to troops at frontline positions in 2024. Gradually, they took over the complex and risky evacuations of the wounded, using special enclosures to protect the soldier being transported. But this year, Lange says, is “the year of the assault UGV.”
Emerging Ukrainian tactics combine UGVs with real-time reconnaissance and surveillance from aerial drones, which discover enemy troops, often under cover of night. The reconnaissance data are then used by remote operators who guide UGVs as they stalk, corner, and shoot to kill. Oleg Fedoryshyn, the head of research and design at DevDroid, another prominent Ukrainian UGV developer, said the ground robots can be controlled from as far as 100 kilometers away using Starlink connectivity, LTE networks, or mesh-networked military radio systems. The UGVs can also carry strike UAVs (uncrewed aerial vehicles), serve as communication relays for drones, or carry and launch communication relay drones that further extend the range of the attack vehicles. The UGV can lurk in position for up to one week without needing a battery charge, Fedoryshyn said, and wait for the enemy to move closer.
“It’s better than to put people there,” he notes. “A guy with a machine gun is always the first target for the enemy.”
The Droid TW 12.7, by DevDroid, is shown here outfitted with a 0.50-caliber M2 Browning machine gun that can be aimed and fired by a remote operator using a tablet and an encrypted communications link.DevDroid
Fedoryshyn estimates that UGVs could eventually help cut the number of soldiers needed along the front line by 30 to 40 percent. Drozhak is even more ambitious. He envisions a future front line that’s entirely automated, relying on sensors and other systems that are only occasionally serviced by humans.
A guy with a machine gun is always the first target for the enemy.
“Right now, we need a lot of UGVs because there are people on the front line and we need to deliver supplies to them,” he says. “But we can substitute many of them with sensor systems, servicing robots, and UGVs, and then we will not need that many for logistics. At some point, we could have only robots in the kill zone.”
Ukraine, with a prewar population of around 41 million, has lost over 150,000 fighters in the war since 2022, according to estimates by the Center for Strategic and International Studies and others. Many thousands of others have been mutilated or permanently disabled. Even those who return without physical injuries suffer lasting psychological trauma. Drozhak dreams that a future robot army would put an end to the ability of autocratic regimes worldwide to brutalize their neighbors.
“There will be no need to push people on the battlefield anymore,” says Drozhak, the RoverTech CEO. “Once we achieve that in Ukraine, any country with a decent economy would be able to defend themselves just with technology.”
RoverTech’s Tarantula active-protection system, which uses acoustic and visual sensors combined with AI algorithms to detect approaching killer drones, is the first step in that direction, he declares.
“The future battlefield will rely on networks of robotic sensors and autonomous systems that can continuously monitor dangerous areas, provide early warning, and reduce the need for soldiers to expose themselves to direct threats,” he says. “Human operators will remain responsible for critical decisions, but increasingly advanced sensing technologies will help move people away from the most dangerous positions on the battlefield.”
Why UGVs Are Vulnerable
Militaries around the world were looking at UGVs prior to Russia’s 2022 invasion of Ukraine. But those were quite different, explains Samuel Bendett, a defense analyst at the consultancy CNA. They were larger, more complex, and conceived to operate in smaller numbers. The more compact forms now seen in Ukraine are the result of an evolution that paralleled that of the first-person view (FPV) attack drones. Both needed to be cheap as they don’t last long and small to be less conspicuous. Now, the West is trying to understand the overall role of UGVs in future warfare. So far, in Bendett’s view, the impact of UGVs on warfare isn’t as profound as that of the FPVs and other aerial drones.
“Not every terrain would be applicable to using a UGV,” Bendett explains. “So far, a lot fewer countries are seeking to integrate them into their combat operations than UAVs, which very much democratized the way of enabling short-range to mid-range strikes against adversaries.”
UGVs, he points out, are much more susceptible to communication disruptions than UAVs, while being less suitable for autonomous operations and swarming due to the complexity of ground terrain.
“With UAVs, communication is much easier,” according to Bendett. “There are no interferences between the ground station and the UAV save the distance, Earth’s curvature, and the radio horizon. But on Earth, there’s lots of different obstacles that interfere with radio signals.”
Most UGVs rely on Starlink as the first choice for operator control, but even that comes with problems. Starlink signals are easily disrupted by trees and buildings. And Russia, having been cut off from Starlink, is working hard to find ways to jam the system.
On top of that, Lange says, as UAV autonomy progresses, UGVs could be left behind. The reason is that UGVs are likely to remain dependent on operator communication links for some time and will therefore be vulnerable to enemy UAVs that can’t be stopped by jamming systems that still provide some protection today.
“The low production cost of strike drones will mean that UGVs will have to endure a barrage of strikes that might be too much,” Lange says. “The question is whether you can make UGVs more survivable on the front line both in terms of command and control and the actual survivability of that many strikes.”
Still, he thinks there’s “no path back from UGVs.” The idea of distributing a whole range of tasks, performed in the past by a single large and expensive tank, to a fleet of small, cheap UGVs provides more resilience against the omnipresent drones. Moreover, although many international commentators now say that Russia appears to be losing, the war grinds on—and so does the cat-and-mouse game of lethal innovation.
Login
