IEEE Spectrum
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.CoRL 2026: 9–12 November 2026, AUSTINHumanoids 2026: 6–9 December 2026, SANTA CLARA, CAEnjoy today’s videos! When one robot loves another robot very much, and they’re connected by a wire, they can get together and make a brand new robot! SFW.[ JSK Lab ]What happens when Microducks meet? They start to sing together. Each Microduck syncs with its neighbors over Bluetooth, then they compose a song on the fly, in harmony and in perfect time. There’s no score and no conductor. Two, three, or four ducks are enough for a choir to break out.If Pollen isn’t selling Microducks in four packs for this exact reason, WHAT ARE THEY EVEN DOING!?[ Pollen ]In an internal test, Google DeepMind robotics researchers instructed Gemini 4 Argon to watch a video of a human playing cornhole. Using its video understanding and advanced reasoning capabilities, Gemini 4 built a virtual simulation in MuJoCo and accurately calculated the physics and whole-body control required for Apptronik’s Apollo 2 humanoid to sink seven consecutive shots in the real world. This Video-to-Sim-to-Real pipeline compresses weeks of manual modeling into a few hours, transforming ordinary videos into an infinite training ground for physical AI.[ Google DeepMind ]As of Oct. 11, 2026, NASA’s Perseverance rover has driven 28.10 miles (45.22 kilometers), becoming the new off-world record-holder for total distance driven by any vehicle on the surface of another world. The previous record of 28.06 miles (45.16 kilometers) was held by NASA’s Opportunity rover. Perseverance reached that mark on the 2,006th Martian day, or sol, of its mission, covering in five and a half years a distance that took Opportunity almost 15.[ JPL ]Let’s not argue about whether these are robots or not and just agree that they’re cool.[ OATs ]If one cute little robot can clean up one block, how many blocks can 40 cute little robots clean up? Science has the answers![ Paper ]Thanks, Keisuke!Before Rosalind Franklin can explore Mars, it has to get off its lander. Engineers are rehearsing that first journey on Earth, using a full-size rover model and Mars-like terrain. The test recreates some of the challenges Rosalind Franklin will face on the Red Planet, from reduced gravity and a 20-degree incline to uneven ground and obstacles beneath its ramps.[ European Space Agency ]From custom part to running robot, the Atlas shop floor is built on precision, rapid assembly, and real-time collaboration. By machining parts in-house and streamlining assembly and repair under one roof, we are driving the manufacturing velocity needed to turn complex designs into walking hardware at scale.[ Boston Dynamics ]What can a quadruped robot do beyond the lab? Deep in China’s Ailao Mountains, DEEP Robotics M20S joined a team of researchers on a biodiversity field expedition—navigating rugged terrain, scouting routes ahead, carrying research equipment, and capturing wildlife in areas that are difficult to access on foot.[ DEEP Robotics ]The Kamino simulation software is a GPU-accelerated physics solver designed for accurate simulation of complex mechanical systems with kinematic loops, such as parallel manipulators, planar and spatial linkages, and multi-limbed robots with coupled joints, without requiring simplified approximations.[ Disney Research ]The original Kawasaki UNIMATE robot as featured in (remember this show?) the MacNeil/Lehrer Report in 1992.[ Kawasaki Robotics ]WildOS enables autonomous robots to perform long-range, open-vocabulary object search in previously unmapped and unstructured outdoor environments by unifying geometric safety with semantic visual reasoning. Through a vision-scored navigation graph and beyond-horizon target localization, the system demonstrates more efficient and robust exploration than conventional geometric or purely vision-based approaches in extensive field experiments.[ ETH Zurich Robotics Systems Lab ]Who needs an AI-generated person when you can have an actual robot that looks exactly like an AI-generated person instead?[ UBTECH Robotics ]The recovery behaviors here are impressive, but the robot is far slower than a human would be, suggesting that the value it generates is primarily in novelty. Which is fine, for now, but novelty wears off, and then what?[ Sharpa ]Here are a couple of good panels from ICRA 2026 in Vienna, including one on “Impact and Return on Investment on Humanoid Robots” and a second on “Advancing Sustainability in Robotics.” [ ICRA 2026 ]
IEEE Spectrum
On October 11, the Perseverance rover completed a drive that brought its total distance traveled on Mars to 28.10 miles (45.22 kilometers). This is a new record for Mars rovers, enabled in large part by Perseverance’s ability to drive with much more autonomy than its predecessors. Over the last five years, Perseverance’s autonomy software has been steadily upgraded, and to learn more about how the record was broken and what the future holds for the rover, we spoke with Mark Maimone, the Perseverance Robotic Operations Deputy Team Chief at NASA’s Jet Propulsion Laboratory, who designed and developed the self-driving software for Mars rovers Spirit, Opportunity, Curiosity, and Perseverance.Before we talk about Perseverance, let’s pour one out for the previous record holder Opportunity, which landed on Mars in January of 2004 for a 3-month mission and ended its epic journey 15 years later in February of 2019. This NASA video compresses Opportunity’s entire mission into four minutes, and gives a poignant sense of the incredible distance that this little rover traveled over its 15 year life on Mars, doing science as it went: NASA Jet Propulsion Laboratory/YouTube Opportunity drove 45.16 kilometers until a planetary-scale dust storm stopped it in its tracks, appropriately enough, in Perseverance Valley.On Feb. 18, 2021, Perseverance joined NASA’s Curiosity rover on the red planet, carrying the Ingenuity helicopter tucked under its belly. Curiosity and Perseverance both ditched Opportunity’s solar panels for a radioisotope thermoelectric generator. Curiosity has been active on Mars since 2012.Despite Curiosity’s nearly 10 year head start, Perseverance is the rover to first break Opportunity’s record. There are several reasons for Perseverance overtaking Opportunity—ruggedized wheels for one—but the biggest contributor is that Perseverance was built for autonomy. To understand how JPL taught a Mars rover to drive itself from 225 million kilometers away without crashing, IEEE Spectrum spoke with Mark Maimone, the Perseverance Robotic Operations Deputy Team Chief at NASA’s Jet Propulsion Laboratory. This animation shows Perseverance’s point of view during a drive of 807 feet (246 meters) along the rim of Jezero Crater on Dec. 10, 2025, the 1,709th Martian day, or sol, of the mission. Captured over two hours and 35 minutes, 53 Navigation Camera (Navcam) image pairs were combined with rover data on orientation, wheel speed, and steering angle, as well as data from Perseverance’s Inertial Measurement Unit, and placed into a 3D virtual environment. The result is this reconstruction with virtual frames inserted about every 4 inches (0.1 meters) of drive progress. JPL-Caltech/NASA Let’s talk about rover autonomy on Mars. Even before Perseverance, rovers were driving themselves, right?Mark Maimone: Yes! Sojourner [in 1997] had an early kind of autonomous behavior—for sensing, it used four laser stripes projected onto the surface, which it used to measure the height of 20 points in the terrain in front of it. It could steer around big obstacles, or even thread the needle between two things. It was a real autonomy system, which is very impressive considering its CPU was less than one megahertz. With Spirit and Opportunity, I developed the autonomy for those rovers. Every time they drove, they would use 256 x 256 pixel stereo pairs to map the geometry of the terrain and find a safe route forward. Spirit actually ended up driving itself 23% of the time, which I was frankly shocked by, since it drove much more slowly in that mode. But using autonomy helped Spirit travel 50% faster than it otherwise could have. Opportunity and Curiosity both only used autonomous driving about 5% of the time. Perseverance is the first Mars rover that drives itself 90% of the time, with its enhanced autonomy system that includes a second processor with an FPGA to do speed up extra image processing operations.What was it like to let Perseverance drive itself for the first time?Maimone: Perseverance is the culmination of decades of work on Mars rover autonomy, so seeing Perseverance use its AutoNav system to drive itself basically at full speed, up to 120 meters an hour, was incredible. These two images taken by Perseverance’s hazard avoidance and navigation cameras show (on left) the very first AutoNav drive on Mars where the rover autonomously stopped before running into a rock pile, and (on right) the second AutoNav drive where the rover successfully steered around a rock.JPL-Caltech/NASAWhere are the humans in this process?Maimone: The plan of where to go is specified by the science team, and then on the engineering side, the strategic route planning team develops a path to get us to that goal by guessing, based on lower-resolution maps from orbit, what the terrain is going to look like for the rover. Each day a tactical operations team combines the strategic overall route with that day’s science goals to pick the actual route the rover will follow.We want humans to keep doing long-range planning because that avoids wasting time. For example, we’ve learned that if we see sandy areas with intersecting waves or crests—not just ripples blown in one direction, but cross-ripples going in multiple directions—the rover tends to sink into terrain like that, so we tell the rover to keep out of those areas. The onboard autonomy only measures terrain shape, not terrain type, so it benefits from those location-based annotations.We now have years of experience correlating images taken from orbit with images taken by the rover, and understanding how different kinds of terrain affect the drive. With the current orbiters at Mars, we sometimes get 25 cm resolution on orbital images. That’s still not quite enough to see all obstacles; there could be a big rock that doesn’t really show up on a scale of more than one or two pixels, making it very hard to notice. That is why you want to have autonomy on the rover, but the greater help we can give it at a high level, the faster we can reach our goals and the fewer potential delays we’ll encounter. This orbital view shows a region of complex Martian terrain that Perseverance could not navigate autonomously. The blue line shows the path that human planners estimated, while the white line shows the path taken by the rover. The inset image, from Perseverance’s hazard avoidance camera, shows the rocky, sandy, and sloped terrain.Source images: Olivier Toupet, Masahiro Ono, et al.; JPL-Caltech/University of Arizona/NASA What happens when the autonomy system on Perseverance fails?Maimone: We ended up not using the autonomy at all for about two months in more complex terrain while we tuned things and wrote new software to deal with that new terrain we hadn’t encountered before. Some of those changes were relatively quick, and some required months of study. For example, one parameter is the maximum amount of slope that the robot is allowed to climb, because on steep slopes, there is a risk of the rover slipping sideways. We landed inside Jezero Crater, which was very flat, and so for the first three years most of our driving was at less than 14 degrees of slope. When we finally decided to leave the crater, we began to encounter other terrain at higher tilts with greater rock density, and the experience we gained allowed us to re-tune the parameters to allow the rover to traverse those more difficult terrains, with slopes of up to 24 degrees.Rover autonomy in the age of AILate last year, you let Anthropic’s Claude AI plan a 400 meter drive for Perseverance. How did that go?Maimone: We had the AI look at the orbital images and pick out a path for us, which became the starting point of the day’s rover planning operations. Typically, that starting path comes from the human rover planner team, but in this case, it was generated by AI. Humans were still in the loop; the team reviewed it, and made a couple of adjustments. AI gave us a solid starting point that allowed us to make rapid progress, because it’s much easier to validate a path than to pick one from scratch. NASA’s Perseverance Mars rover took this selfie on July 23, 2024. The selfie is composed of 62 images taken by the WATSON (Wide Angle Topographic Sensor for Operations and eNgineering) camera on the end of the rover’s robotic arm.JPL-Caltech/MSSS/NASADo you think humans will remain a necessary part of rover operations, or do you see the next generation of Mars rovers operating for weeks at a time without human assistance?Maimone: One of the things that has held us back from driving very long distances autonomously is that as we drive, the rover accumulates uncertainty in its position. When the uncertainty grows, it means we have to stay farther away from areas that human rover planners have identified as being potentially dangerous, called ‘keep-out zones.’ In some cases, the uncertainty gets high enough that the rover can’t plan a safe path, because the keep-out zones (which grow with the rover’s uncertainty) get too big.We now have a new capability that we’ve added to Perseverance called Mars Global Localization. With this new technology that can shrink our uncertainty, we theoretically have the ability to go as long as we have resources to keep driving safely based on our maps and keep-out zones.But to truly run a mission for weeks, you need to not only drive long distances, but also execute science operations autonomously. We have people here also looking at automating science operations by doing image processing onboard the rover to get a better understanding of what’s out there and where to focus the instruments’ attention. So I do think we are getting to the point where we can support long-period autonomous operations. It has been a slow evolution, but we have great capabilities now and a good understanding of what’s needed to make it work well.As someone who has been involved with all of these Mars rovers—including one that got stuck—do you fundamentally trust Perseverance to operate autonomously?Maimone: I do. We built the software, tested it, ran millions of simulations, and did lots of work in the field in many different test environments. It hasn’t steered us wrong yet. We’re looking forward to setting even bigger records.
IEEE Spectrum
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.CoRL 2026: 9–12 November 2026, AUSTINHumanoids 2026: 6–9 December 2026, SANTA CLARA, CAEnjoy today’s videos! It’s possible that I wasn’t the only one who thought that Figure’s recent stunt was somewhat ridiculous.Devanthro is in fact selling these plastic balls for $1000 each, and they’ll arrive in time for Christmas.[ Devanthro ]Not that anyone uses ringtones anymore, but if we did, this would be absolutely everywhere and I’d be loving it.[ Pollen ]Let’s not get too excited, but this thread-like linear motor is pretty sweet.[ Ecole Polytechnique Fédérale de Lausanne LMTS ]What is this robot for? It’s for doing exactly what you see here, that’s what it’s for.[ RAI Institute ]Humanoid robots still struggle to plan contact-rich whole-body manipulation from egocentric RGB and proprioception. Workhorse learns such manipulation from robot-free human demonstrations. On a real Unitree G1, Workhorse sorts boxes with its hands and a kick, catches a thrown box, and topples and climbs a suitcase.Yeah that happens to me too when I have a communication failure.[ Hybrid Robotics ]A car wash that models your vehicle then cleans it with robot arms? Interesting idea.The recent reviews from this location suggest that the robots aren’t quite up to snuff yet, but really seems like there’s potential here.[ KUKA ]So this is not the worst idea I’ve ever seen for what to do with humanoid robots.[ DEEP Robotics ]Some lovely manipulation with a deceptively simple gripper.[ Generalist ]I am genuinely embarrassed to say that this robot dances better than I do.[ General Robotics Lab ]MIT researchers developed SANDO, a new navigation system that helps robots safely navigate unfamiliar environments while avoiding moving obstacles. The system can rapidly plan and adjust a collision-free path using onboard sensors and computing, with safety backed by mathematical guarantees.[ MIT News ]Can you spot the part of this video that I’m side eyeing?Yep, it’s where the middle clip starts after the hand already has a good grip on that peelable strip.[ Sanctuary AI ]Inspired by feather stars, the NC State researchers report a soft underwater swimming robot capable of achieving three distinct swimming modes—jellyfish-like vertical swimming, fish-like directional steering, and rotor-like 3D reorientation—using only two pneumatic inputs.[ NC State University ]Here’s a series of mini-keynotes from ICRA earlier this year, on manipulation, humaonids, and embodied design.[ ICRA 2026 ]
IEEE Spectrum
A hand attached to the end of an arm is a given for most humans, but what if robots didn’t have to play by the same rules as we do? Researchers from ETH Zurich have turned an off-the-shelf robotic hand into a completely self-contained robot that walks on its fingertips while using those same fingers to manipulate its environment.Besides making the perfect Halloween accessory for anyone planning to dress up as a member of The Addams Family – who were famously accompanied by a disembodied walking hand called Thing – the researchers say the approach could make it possible to operate controls and handle objects in places a traditional robot arm can’t reach. Soft Robotics Lab, ETH Zurich/YouTube Amirhossein Kazemipour, a Ph.D. candidate in the Soft Robotics Lab at ETH Zurich in Switzerland who led the project, believes this hand could prove useful in contexts like fiddly maintenance tasks in industrial settings, or in search and rescue, where a robot could detach its hand to explore confined spaces too tight for the rest of its body. More broadly, he says, the work is a step toward a new approach to robotics in which body parts are not confined to a single role.“I wanted to see how much a robotic hand could do without the rest of the robot,” Kazemipour told IEEE Spectrum. “Instead of having this traditional view of robotic parts, what if we could use those parts to do other tasks that they are not designed to do.”This is not the first attempt to get a robot hand to walk, but previous work has typically relied on specially designed hands. The ETH team instead decided to use off-the-shelf hardware from Wuji Technology, augmented with an 80-gram backpack containing a battery, an inertial measurement unit (IMU), and a Raspberry Pi Zero. Kazemipour says they decided to go with a commercial robot hand because they didn’t want to compromise on the hand’s manipulation capabilities. This introduced significant challenges, he adds, because a hand’s shape is optimized for grasping rather than crawling.How to Walk With FingersTo teach the hand to walk, the researchers trained a model in simulation using reinforcement learning – an approach in which models learn skills through many rounds of trial and error, receiving rewards for getting closer to the desired behavior along with penalties when they diverge from it. But Kazemipour says the unusual geometry of a hand meant they couldn’t just reuse approaches used to train other kinds of legged robots. Standard quadrupeds or bipeds tend to be symmetrical on their left and right sides, which simplifies walking and balance, he says. Hands, in contrast, are not at all symmetrical, with fingers of different lengths and an opposable thumb.In addition to rewards for moving at the right speed and in the right direction and penalties for things like wobbling or jerky movement, the team introduced a novel way to keep the hand in a crawling stance during the simulated training. To account for unequal digit length, they gave each finger its own target position relative to the palm, based on where it naturally sits when in the crawling posture. They then penalized the fingers for straying too far from that reference point, creating what they describe as “virtual springs” that pull the fingers back to the crawling stance. Crucially, the penalties for fingers moving forward and backward are significantly less than those for moving side-to-side, so taking steps in that way costs the hand very little compared with the rewards it receives for moving.The team also trained separate models for skills like righting the hand when it falls over, pressing keys on a keyboard, and pushing objects to a target. All four models fit on the onboard computer, says Kazemipour, and the controller can switch between them as necessary. In real-world tests, the hand crawled across 14 different surfaces including smooth floors, metal grates, grass, and gravel. Other experiments confirmed that it could make both right and left turns, and hit an average speed of 9 centimeters per second. The hand was able to right itself in 21 of 25 trials. When given a video feed from an overhead camera, it even managed to autonomously push a 41-gram cube to targets up to 40 cm away 15 times in a row.“What interests me most about this work is that the same fingers can move the hand, support its weight, and interact with its surroundings,” says Masahiko Inami, a professor at The University of Tokyo whose group has developed hand-shaped walking robots. He agrees that a mobile hand could be useful for reaching confined spaces and operating controls that an entire arm could not easily access, though he says practical deployment would require on-board perception and more robust navigation. Hideki Shimobayashi, a Ph.D. student in Inami’s lab, also points out that walking puts very different loads on a hand compared to grasping, so durability may be a challenge.However, Matei Ciocarlie, an associate professor of mechanical engineering at Columbia University, says it’s a “cool result” that suggests robot hands don’t have to be restricted to the same capabilities as their human counterparts. “One can imagine an entirely new class of mobile manipulators that are simultaneously dexterous, able to traverse complex terrain, and able to manipulate payloads comparable in size to themselves,” he says.Kazemipour sees his work as part of a longer-term vision for robots made up of parts that are not tied to a single job or even a single body plan, something he calls “autonomous modular embodiments.” “Robotic body parts should not be permanently assigned to a single function,” he says. “They can dynamically transition between roles, between being components of one embodiment system and autonomous agents, according to the task.”
IEEE Spectrum
This White Paper gives robotics researchers and engineers an overview of a new large-scale motion capture dataset built to close the data gap limiting humanoid robot learning. It also shows how policies trained on the dataset transfer to a real humanoid robot.What you will learn about: Why humanoid robot learning, a central problem in embodied AI and Physical AI, needs data that internet video and existing motion capture datasets cannot provide. How FrameNet, a linguistic framework for human action, can guide motion capture collection to systematically cover a broad range of whole-body motion. Why synchronized object trajectories and meshes make human-object interaction data useful for teaching robots real-world tasks such as carrying, pushing, and pulling. How reinforcement learning policies trained on this motion capture data improve with scale, and how sim-to-real transfer carries them onto a physical humanoid robot.Download this free whitepaper now!
IEEE Spectrum
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.CoRL 2026: 9–12 November 2026, AUSTINHumanoids 2026: 6–9 December 2026, SANTA CLARA, CAEnjoy today’s videos! We’ve developed ALBATROSS, a hybrid aerial–marine robot that can be released from the air, autorotate down to the water without a parachute, passively self-right after landing, and then reuse the same rigid wings as sails for autonomous wind-powered navigation.The key idea is to make the physical structures do as much of the work as possible. Rather than carrying separate systems for aerial descent, landing, and marine propulsion, ALBATROSS uses the same wings for both autorotation and sailing, with minimal actuation. Ultimately, we’re interested in whether platforms like this could combine rapid aerial deployment with persistent, energy-efficient marine sensing.[ Singapore University of Technology and Design ] paper via [ Science Robotics ]Thanks, Shane!I’m just getting so burnt out on the humanoid publicity stunts, you know?[ Figure ]Sharpa unveiled three major proprietary new products at the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS): D01, the first integrated tactile-sensing robot built for dexterous manipulation; W02, a next-generation fully tactile, ultra-compact, lightweight dexterous hand; and AE01, a high-fidelity haptic exoskeleton data glove.[ Sharpa ]Why will Boston Dynamics win the humanoid robotics race? We’ve already commercialized autonomous mobile robots, creating markets with Spot and Stretch. Now we’re doing it again with Atlas: learning in real environments, designing for manufacturing at scale, and introducing a new era of physical intelligence.[ Boston Dynamics ]This work presents a lightweight, compliant footpad that enhances the terrain adaptability of a hopping robot. The design combines a planar foot supported by a compliant spherical joint for self‑alignment with embedded spines to improve traction on uneven or slippery surfaces.[ CLIMB Lab, University of Toronto ]Heterogeneous robot teams distribute complementary capabilities across specialized agents, but their physical roles and capacities typically remain fixed throughout a mission. We present HARP, a Heterogeneous Aerial Robotic modules Platform in which independently deployable aerial robots physically reconfigure to compose their capabilities for field operations.[ General Robotics Lab ]Some very impressive mobility from ANYmal.[ ETH Zurich RSL ]Daniel writes, “I’m a high school student in Canada, and I built a self-constructing robot arm that assembles itself. It’s designed to be low-cost, with disaster response in mind: small robots can squeeze through gaps in rubble but are too weak to move things, while big arms are strong but can’t fit through. My arm is split into three wheeled modules that travel separately, then dock together using a spring-lock mechanism.”[ Daniel Zhu ]Autonomous highway driving not impressive enough for you? Try this.[ XPENG ]The IT Imperial team has connected MaleCNS—a published connectome of a male fruit fly (Drosophila melanogaster)—to a physical Unitree G1 humanoid robot. The system creates a closed loop: camera → fruit-fly neural model → motor decoding → physical robot movement. The robot receives visual input through its stereo camera. The images are transformed into a representation adapted to the fly’s visual system, processed by the computational model based on the reconstructed nervous system, and then converted into movement commands for the robot.[ IT Imperial ]Thanks, Anna!My guess is that this is far easier than it looks or far harder than it looks and I’m honestly not sure which.[ Paper ]I cannot believe that this is still a thing, but good on them![ Tevel ]In Part 1 of Dexterity Explained, we break down what manual dexterity is, why it matters, and how robotic hands and Physical AI work together to bring greater skill, precision, and control to physical tasks.[ Sanctuary AI ]XPENG has a lot of money to build a lot of these robots. They do look very human. I’m not sure what they’re going to do, though?[ XPENG ]That’s a lot of robots! Tell me what they’re all going to do![ UBTECH ]When floods swept away roads and bridges in Nepal, local drone pilots stepped in to help map the damage, support search teams, and deliver essential supplies to communities cut off from help. In this episode of Ground Truth with Flying Labs, host Leka Tingitana speaks with Uttam Pudasaini, an advisor to Nepal Flying Labs, and Raj Bikram Maharjan, co-founder of Airlift Technology, about the drone community’s response to the devastating floods of 26 August 2026.[ WeRobotics ]
IEEE Spectrum
Today, Boston Dynamics is announcing a redesigned hand for its Atlas robot. The previous generation could do some crazy superhuman things with its three fingers, but it was never intended for mass production in the tens or hundreds of thousands. Switching from research hardware to a scalable product makes previously ignorable questions suddenly paramount: how can you make a hand that does everything you need it to do, but is also rugged and reliable and cost effective? It’s questions like these that drove the design of the new Atlas hand, and while it doesn’t look much like a human hand, it might be a hand that makes it out into the real world. - YouTube www.youtube.com Compared to the almost eerily human-like hands designed by a litany of other humanoid robotics companies, Atlas’ new hand does look awfully clunky. We’ve certainly seen some of those robots with much more human-like hands perform dexterous manipulation tasks that are, speaking as someone who has personally witnessed robots haplessly fumble their way through the world for nearly two decades now, nothing short of incredible. And it’s not just what these hands can do—it’s how they look while doing it, as slim, graceful mechanical copies of our own hands. They seem like science fiction, and to some extent, that’s exactly what they are.“Hands are a ruthless design tradeoff,” Alberto Rodriguez, Director of Robot Behavior at Boston Dynamics, tells IEEE Spectrum. “There’s no way around it, you’re always giving up on something. And many of the current designs are giving up on things like reliability and manufacturability.” This is what invariably happens when you try and cram everything necessary to make a robotic hand function into the form factor of a human hand: it’s going to be either very fragile or very expensive, and likely both at the same time. The unfortunate reality is that for now, and likely for the immediate future, all that these sorts of hands are good for are research and demos.With Atlas’ new hand, Boston Dynamics is making the design tradeoffs necessary to achieve a hand that, according to the press release: is capable of a vast array of useful tasks, including using tools; is strong and rugged; can be cleanly simulated; can be mass manufactured and repaired at reliably low cost, and is close enough in form factor to a human hand that human demonstrations can be used to train it.You’ll notice that “looks cosmetically like a human hand” is nowhere on that list, and this new hand doesn’t. Rodriguez explains that getting rid of the pinky was a straightforward decision, agreed on after his team spent a day with their pinkies taped to their ring fingers. But why even add a fourth finger, when the previous generation of Atlas’ hand seemed to do well with only three? It’s not just the extra finger, but a complete redesign that almost doubled the degrees of freedom between the old hand and the new, going from 7 to 13. Boston Dynamics“We also added the ability for the fingers to splay open, because we knew that it was important for one finger to be able to move against another, and we also knew that we needed to be able to hold the handle of a tool with a trigger,” Rodriguez says, which includes many common tools like drills or welding torches. “We’re convinced that from a product perspective, this hand should be able to do everything we need it to do for the foreseeable horizon.” And that ability to finger splay goes beyond what human hands are capable of. According to Rodriguez, “with reinforcement learning we can discover uses for these superhuman extra motions, and exploit them.”Reliability, manufacturability, and cost are all closely related. By reducing the degrees of freedom relative to other more anthropomorphic hand designs down to just 13, Boston Dynamics can use fewer, larger, and more powerful actuators. The actuators are embedded directly into the joints themselves in a direct drive configuration, with a transmission that allows the motors to be back-driven and react to force and contact. Each actuator pack is a single unit that can be easily replaced, and there are no delicate tendons or cables across joints to stretch and break. “What we’re working on right now is figuring out what needs to change in the fine details of our design if we want to make 100,000 of these hands a year,” Rodriguez says.Not every humanoid company will make the tradeoffs that Boston Dynamics has done here, and there’s still plenty of room for creativity when it comes to hand design. But if humanoids with hands are going to scale, some difficult decisions will have to be made, and it’ll take some compromise to deliver on the promises that the humanoid industry has committed to.
IEEE Spectrum
Building useful robots starts with understanding the people who use them. For Charlie Kemp, cofounder and chief technology officer of Hello Robot, that means developing assistive robots that can help people with everyday tasks and support greater independence.In this Robots Guide profile, Kemp shares his path from studying artificial intelligence at MIT to building Stretch, explains how working with people with disabilities has shaped his approach, and offers advice for aspiring roboticists. Read the full profile on IEEE’s Robots Guide.
IEEE Spectrum
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.IROS 2026: 27 September–1 October 2026, PITTSBURGHCoRL 2026: 9–12 November 2026, AUSTINHumanoids 2026: 6–9 December 2026, SANTA CLARA, CAEnjoy today’s videos! Life’s better with a little Goose.[ The Robot Works ]Thanks, Dave!Skydio has introduced the F10, a charmingly lopsided fixed-wing drone, that uses a robot arm for autonomous launch and capture.Unfortunately, based on the frequency of that flashing LED on the drone, this video appears to have sped up the capture (and likely the launch) by a substantial amount.[ Skydio ]I want a bunch of these little guys. I don’t know what I’d do with them. But I want them.[ Freeform Robotics ]Walking robots are everywhere now… But 30 years ago, they were brand new. The first ever robot that could walk by itself was the Honda P2. The P2 featured motors and hydraulics in the hips, knees, and ankles to walk around, and weighed over 300 pounds.[ IEEE Spectrum ]Precision biomimetic dexterous hand, priced from US $6,500.[ Unitree ]What if all it took to train a drone to do acrobatics was to tell it whether you like the way it flies?[ UZH Robotics and Perception Group ]Why build a humanoid hand when a robot hand can do stuff like this?[ General Robotics Lab ]Diane Heinle, a mechanical engineer at Boston Dynamics, answers questions about endurance testing for Stretch, how her team simulates years of wear on robot parts over a brief time in the lab, and why there are bird heads on them.[ Boston Dynamics ]ULOHA brings learning from demonstrations to two robot arms working underwater. We developed custom leader–follower hardware and software extending LeRobot to connect teleoperation, demonstration collection, and autonomous manipulation. The video shows the robots lifting objects together, passing objects between arms, opening a container, and catching a sponge as it rises through the water. We also explore how bubbles and action-execution timing affect learned behaviors.[ ULOHA ]Thanks, Masato!Usually, ballbots rely on an internal pendulum, but here’s a different approach.[ General Robotics Lab ]From long-horizon manipulation to tactile-guided actions, this video highlights selected collaborations that explore what Astribot robots can do.[ Astribot ]This is one of those tasks that looks a little boring but is actually both very tricky and very useful.[ HiPeR Lab ]That tiny little excavator playing in flour is adorbs.[ ETH Zurich Robotic Systems Lab ]How do you fit a humanoid robot into a box?[ LimX Dynamics ]And now, this.At the Opening Ceremony of WorldSkills Shanghai 2026 on 22 September, 19 Unitree humanoid robots performed alongside 120 dancers, presenting the world’s largest-scale performance featuring full-size general-purpose humanoid robots before an audience of more than 10,000 people, with a fully AI-driven autonomous robot cluster performance live-streamed worldwide in real time.[ Unitree ]A microwave box creates a surprisingly tricky vision system challenge: there’s a picture of a microwave on the side of the box, and that microwave has a window. From the camera’s perspective, it’s a box inside a box inside a box.[ Pickle Robot Company ]What exactly is Physical AI? Personally I think it’s just what robotics has always been, more or less, except that everything has to have “AI” in it now so here we are.[ MIT CSAIL ]As the evolution of the UR Series, the g‑Series is designed to accelerate how robots are integrated into complete automation systems. It connects more directly with cameras, higher power EOATs, and control systems, eliminating external cabling and hardware to streamline integration. This enables cleaner and faster deployment of advanced automation.[ Universal Robots ]
IEEE Spectrum
In Guadalajara, Mexico, many high schools have motivated teachers and talented students with an interest in science, technology, engineering, and mathematics, but they lack access to advanced tools such as robotics laboratories. The resources shortfall limits the students’ opportunities for hands-on learning on cutting-edge applications.A team from ITESO, Universidad Jesuita de Guadalajara, is working to change that. Through the EPICS in IEEE initiative, a multidisciplinary group of 15 engineering students, faculty advisors, and IEEE Guadalajara Section volunteers developed RoboMeshA. The portable, self-contained educational platform brings robotics and AI experiences into classrooms.EPICS is administered by IEEE Educational Activities and funded by the IEEE Robotics and Automation Society.A mobile laboratoryRather than requiring a school to build a dedicated computer lab or install complex software, RoboMeshA operates as an all-in-one mobile learning network.“RoboMeshA brings robotics and AI to students who don’t have access to specialized facilities or preinstalled software,” says team member Fernando Vidal Luna, an IEEE student member and a mechatronics engineering major at ITESO.Students connect directly to the platform from a user-friendly web browser. They can interact with the robot manually or use its control modes to watch it move and detect and avoid obstacles.“The project combines mechanical design, embedded systems, control engineering, computer vision, and AI into a single robotic system that functions as a mobile learning laboratory,” says faculty advisor Jorge A. Lizarraga.The team says young students are interested in technology, programming, and robotics but don’t have an opportunity to work with systems that combine mechanics, electronics, software, and control.“RoboMeshA allows students to see how all these disciplines work together in a tangible and understandable way,” says team member José S. González, who also is studying mechatronics engineering.The team has built two units and is developing a modular coupling framework to expand the system’s capabilities for research and classroom demonstrations. The structured system design approach connects independent software components while minimizing internal dependencies, enabling four RobotMeshA robots to operate together.Overcoming design challengesThe team faced significant hurdles while designing the project.“One key challenge involved the robot’s structural design,” Luna says. “It wasn’t only about making a chassis where all the components fit and the design had sufficient stability, rigidity, and weight distribution. It was also about ensuring that the electronics were protected while still being accessible for maintenance, testing, and modifications.”“It was also challenging to design a platform that could be used by students with different levels of experience,” González adds.“When students realize the technology they develop can inspire others and improve lives, engineering becomes far more meaningful.” —Luis Fernando Luque-VegaThe team partnered with the CETI Colomos and Prepa ITESO high schools to validate the platform in classroom settings.“We wanted the first interactions with the robot to be simple and intuitive,” González says, “such that students could simply power the robot, connect to its network, and begin interacting with it, rather than having to deal with software installation, extensive configuration, or troubleshooting.”Engineering with social impactMany of the students who participated were from ITESO’s applied professional projects program. The experience offered them practical training in project management, system integration, and user-centered design.The team also presented a research paper and a project poster in May at the Engineering Congress of the Jesuit University System.“Seeing a design move from a digital model to a physical system was invaluable,” González says. “Working with students from different backgrounds taught us to listen to end users and design for their actual needs.”Project lead Luis Fernando Luque-Vega, an IEEE member, says he’d like the venture to serve as a blueprint for engineering education.“I hope RoboMeshA is adopted by schools, universities, and IEEE student branches across Mexico and internationally as a model for integrating technical innovation with community engagement,” Luque-Vega says.By pairing engineering talent with community service, initiatives such as EPICS in IEEE demonstrate how targeted support can turn academic concepts into real-world solutions.“When students realize the technology they develop can inspire others and improve lives, engineering becomes far more meaningful,” Luque-Vega says.For more information on service-learning opportunities, visit the EPICS website.