Shift Robotics launched the Moonwalkers Dusk this week. Powered shoes that strap over the ones you're already wearing, add force to each step, and get you to 7 mph without breaking into a run. $1,199, shipping late October. The headline is the gait model. Their S.E.N.S.A. system learns your stride inside about 10 steps, tells walking from running from jumping, estimates the slope you're on and adjusts assistance to match. Foot gestures switch modes so you're not pulling out a phone. 3.5 to 4.2 lb per foot depending on size, 90-minute charge, 6 to 7 mile range, 220 lb rider limit, IP54 — splash-resistant, not waterproof. But the line in the release that caught our eye at RobotAutoTrader is much further down. The straps, plastics and wheel rims come off without tools. That sounds like a footnote and it isn't. A consumer robot keeps its value when a second owner can replace the parts that wear out. The moment a cracked plastic or a worn wheel means a service centre, resale collapses — because a buyer is being asked to pay for a machine with unknown remaining life and no way to restore it. Wheels wear out. Batteries fade. Whether that's the end of a $1,199 machine or the beginning of its second life mostly comes down to whether you need a screwdriver.

Figure AI destroyed most of its F.02 fleet this week. The robots were trained to walk to the lip of a 75-ton electric arc furnace in Imatra, Finland & jump in on their own. 😢 Six melting cycles in 24 hours. It was actually Schwarzenegger's idea & he showed up for it. Finland because no foundry in the US or Mexico would let them drop a machine full of lithium-ion batteries into expensive equipment. But that's just the logistics. The reason they were destroyed at all is that Figure wanted it that way: a disassembly would have eaten enough engineering time to delay the F.04 launch & the company wanted its proprietary hardware gone rather than sitting in someone else's workshop. That's the part worth sitting with. These machines weren't broken. A manufacturer looked at a fleet of working humanoids & decided the safest outcome was for them to stop existing. We track used robot values at RobotAutoTrader & this is the hardest version of the problem. Not a thin market, but a market the maker actively prevented. Every other category of machine eventually gets a second owner & that second owner is why the first one can finance the purchase at all. Figure is now machining the recovered steel into limited-run commemorative artifacts and selling those. So the metal gets a resale market. The robots don't. 🤖

Four Growers just put a FANUC robot arm to work picking tomatoes in the Netherlands. A greenhouse in the country is now using the AI-powered system to identify ripe tomatoes and pick them. Four Growers developed it, pairing the AI with the FANUC arm. The work splits into two parts. The AI decides which tomatoes are ready. The arm then does the physical picking, with the precision to harvest without damaging the fruit. Both parts have to work, because a robot that misjudges ripeness or bruises the crop saves no one any time. The system reduces the greenhouse's labor needs. The people who work there can spend more of their time on plant care instead of repetitive harvesting. The robot takes over the picking, and the workers keep the tasks that need a grower's attention. An industrial arm doing delicate agricultural work shows how far automation has moved beyond the factory floor. As machines like this spread through greenhouses, there is still no vertical second-hand market where a used one can be resold with its hours & condition on record.

Hyundai plans to put more than 25,000 Atlas humanoids into its own factories. Boston Dynamics opened the training centre for them inside the Georgia Metaplant last week. The numbers underneath that are the real story. Hyundai is targeting 30,000 Atlas units a year by 2028, and roughly 83% of that output goes into its own plants before anyone else can buy one. Hyundai Mobis is building a US actuator facility with capacity for 350,000 units a year — actuators being the joints, and around 60% of a humanoid's material cost. Each Atlas needs more than ten. That's not a pilot. That's a supply chain. And it changes what the second-hand market looks like. Everything until now has been ones and twos — a lab selling a spare, a startup winding down. A fleet of 25,000 units running production shifts creates something different: standardised hardware, documented service intervals, a real parts supply, and eventually a refresh cycle that retires thousands of working machines at once. That's exactly how used cars became a functioning market. Fleets first, individuals second. The other thing 350,000 actuators a year buys you is repairability. When the most expensive component in a robot is being mass-produced domestically, fixing an older unit stops being a specialist job and starts being a parts order. The second-hand humanoid market doesn't exist yet. When it does, someone needs to have been tracking what these machines are actually worth. That's what we're doing with Robot Autotrader.

Nvidia is talking to insurers about a product that doesn't exist yet: insurance against its own chips losing value too fast. The structure, per the Financial Times, is residual-value cover. A smaller cloud operator borrows against its GPUs. If it defaults and the chips can't be resold for enough to cover the loan, the insurer eats the difference. Nvidia is working with reinsurance broker Howden Re, and has shared chip depreciation data and forward compute-price models with at least one insurer. The comparison Jensen Huang keeps making is aircraft. Planes are financed against residual value because the industry has decades of resale data — everyone knows roughly what a 12-year-old airframe fetches. GPUs have almost none of that. The chips are new, the refresh cycles are fast, and nobody can say with confidence what an H100 is worth in 2029. That's the whole problem. You cannot underwrite a depreciation curve that has never been measured. GPUs have almost none of that. The chips are new, the refresh cycles are fast, and nobody can say with confidence what an H100 is worth in 2029. That's the whole problem. You cannot underwrite a depreciation curve that has never been measured. Which is why Nvidia is handing over its own depreciation data to get this off the ground. Before capital will finance hardware, somebody has to establish what used hardware is actually worth. That's the same problem one tier down, in robotics. A warehouse operator financing twenty units, a lab upgrading a fleet, an owner deciding whether to sell — all of them are guessing, because nobody's been tracking what these machines resell for. It's why we built the Robot Resale Index. 1,140 tracked sales across 62 models, currently at 96.4 & down 3.6%. Not a marketing number , it’s the beginning of a depreciation curve. Aircraft got financeable when their resale value became measurable. Robots will go the same way.

Four years ago a Tokyo lab grew living human skin onto a robotic finger. It healed itself when cut. It also couldn't survive without a constant supply of nutrients and waste removal. That second part is why you haven't seen it since. The work is real & genuinely clever. Rather than growing skin separately and gluing it on, the team dunked a three-jointed robotic finger in a solution of collagen and human dermal fibroblasts. Those cells naturally contract as they culture, so the tissue shrank and wrapped itself tight around the finger — 1.5mm thick, with proper dermis and epidermis layers, moving with the joints without tearing. Cut it, apply a collagen patch, and the cells knit it in within about a week. The problem is that living tissue is a maintenance contract. It needs feeding, it needs waste removed, it dies if sterility slips & it's weaker than the skin you're wearing. Which is the useful lesson for anyone buying machines. The most impressive capability in a spec sheet is often the one with the highest upkeep. Self-healing sounds like less maintenance and is actually more. The robots that hold their value are the boring ones — sealed, serviceable, parts you can still buy in five years. That's what we check on every listing at Robot Autotrader, because the thing that fails first is rarely the thing in the headline.

What to check before you buy a used.🤖 5 things decide whether a second-hand robot is a bargain or a brick. 🔋🪫 Battery health, measured not guessed. Lithium-ion packs lose capacity on a fairly predictable curve. A robot vacuum that ran 90 minutes new might manage 45 after three years of daily charging. Ask for actual runtime on a full charge, not the box spec. Then check whether a replacement pack is even sold separately & what it costs. On some models the battery is £40 and ten minutes. On others it isn't user-replaceable at all. 📲 Whether it still gets firmware. This is the one that catches people. Plenty of robots depend on the manufacturer's servers for mapping, scheduling or app control. When a company drops support or goes under, features quietly disappear from a machine that's still physically perfect. Check the current firmware version and whether the model is still supported. 🛞 Parts availability. Brushes, filters, wheels and side modules are consumables. A robot with a healthy spares market stays useful for years. One whose manufacturer stopped making parts is on a countdown, regardless of condition. ⚙️ Wear on the moving bits. Drive wheels flatten, brush bearings seize, mower blades dull, LiDAR turrets get sticky. These are cheap to fix if you know before buying and annoying if you find out after. 📝 Who owned it and why they're selling. An upgrade is a good reason. A persistent fault is not. Ask what the error history looks like and whether the dock and accessories are included as docks are surprisingly expensive to replace on their own. Robots don't have odometers, which is exactly why this information has to come from somewhere. At Robot Autotrader we put this on every listing so you're not guessing.

Nineteen humanoid robots waltzed with human partners at the WorldSkills opening ceremony in Shanghai, and the livestream audience couldn't reliably tell which dancers were which until some robots removed their costumes. The machines are Unitree H2s — 180 cm, about 70 kg, 31 degrees of freedom, with a bionic face whose eyes are cameras. Leg joints deliver up to 360 newton-metres of torque, and onboard compute scales to Nvidia's Jetson AGX Thor at 2,070 TOPS. Three-hour battery. Unitree says the routines ran autonomously in real time, with no remote control. Worth being precise: autonomous here means executing learned choreography and holding sync without a pilot, not improvising. It's still a hard problem — close-contact partner dancing means constant balance correction against a moving human who doesn't follow a script exactly. The reason this matters commercially: every one of those 19 robots is a production unit, not a prototype. Unitree ships in volume, prices aggressively, and has iterated H1 to H2 in under two years. Which is exactly the pattern that creates a resale market. Fast iteration means the previous generation is still fully capable when the new one lands — the H1s that danced at the Spring Festival Gala didn't stop working when H2 arrived. That's the part worth tracking. Our Robot Resale Index sits at 96.4, down 3.6%, and the reason values soften is supply: upgrades push working machines into second-hand circulation. Falling prices aren't a problem if you're buying & reselling on robotautotrader.com

A robot taught itself to play football by playing against itself for 140 years. Skild AI ran it in simulation, then loaded the result into a real body. The clip is autonomous, real time, no pilot. That's a brain, not a machine. And it's the thing worth understanding if you're buying robots. Skild's approach is one model that works across bodies — humanoids, quadrupeds, arms — trained across thousands of morphologies so it doesn't memorise any single piece of hardware. Their demos include robots adapting to a stuck wheel in seconds and walking with a broken leg after a few tries. Which changes what "outdated" means. When the intelligence lives in a model that keeps improving and transfers across bodies, last year's frame isn't obsolete. It's a body waiting for a better brain. So the question when you buy used stops being how old it is, it becomes: what's the battery actually measuring at, what firmware is it on, can you still get parts, and who owned it. All this will be available at Robot Autotrader before you commit to buy.

Meet Predator. Steel-built, six feet of it, and the reason REK billed it as the Final Boss. Frankie LaPenna put Predator on the floor several times this week. Predator got back up every single time, fast & won the fight anyway. The knockdowns are the whole demo. Standing up after a fall is one of the hardest problems in humanoid robotics — the machine loses its balance reference completely, has to work out its own orientation from its sensors, then drive dozens of joints through a push-up while its weight shifts underneath it. Not long ago that meant a harness and a technician walking into the ring. Predator did it over and over, with a person actively trying to keep it down. That's the property that decides whether a robot can work anywhere near people. Falls stop being downtime. And a frame that absorbs that kind of punishment and keeps standing is a frame that's still earning three years from now. Robots like Predator don't wear out. They get replaced. RobotAutoTrader is where the replaced ones find their next owner.

Unitree just started selling a robot hand on its own. $6,500, 22 joints, which is the same size as yours! Every joint is backdrivable -push a finger and it gives instead of fighting you & each one has impact protection built in. Roboticists are calling it much closer to a human hand than anything at this price. But the real news is that it's sold separately. Hands are the hardest-working part of a humanoid and the part that gets replaced most. They get jammed, dropped, outgrown when a lab switches platforms or a newer model ships. Once a hand has its own price tag, it stops being locked inside one robot. It becomes something you can own, swap & sell. Think about who's on each side of that. A research lab upgrading to next year's hand has a working $6,500 component sitting in a drawer. A student team or small startup would kill for one at a fraction of that. Right now those two people have no good way to find each other. That's the gap we are building for with RobotAutotrader.com Robots don't rust & don't rack up mileage. A hand that worked on day one still works when its first owner moves on….it just needs its second owner!!! Sell the one you replaced. Buy the one someone else already broke in. 🦾

A human boxer just fought a six-foot humanoid in a cage in San Francisco. 🥊 The robot is an EngineAI T800, named after the Terminator, first one imported to the US. It wasn't brought here for a factory or a lab. It was brought here to fight. REK's robots are piloted. Humans wear VR headsets & combat controllers and drive the machine in real time. So the fight everyone's calling man vs machine is closer to man vs man wearing a headset. Real Steel, not Skynet. But the part we care most about is that REK is opening a permanent storefront in San Francisco where you can watch robots, take classes on walking & programming them, rent one, buy one, or bring your own in for repairs. That's the whole lifecycle showing up in a single retail space & it's the clearest sign yet that humanoids are becoming things people own, break, upgrade & resell rather than equipment that lives in a factory. Every category that gets a repair shop eventually gets a used market & this is exactly what Robot Autotrader is getting ready for.

UBTECH's new factory in Liuzhou is designed to build a humanoid robot every 10 minutes, using humanoid robots. Cruzr units unload materials, stack them, and move them between stations while collaborative arms handle final assembly. 14,000 square meters, built with Siemens, running on a digital twin they call the smart brain. Walker S and Cruzr series, a stated capacity of 10,000 units a year. The finished robots go into an automated warehouse holding 112 of them in 65 square meters. A building full of humanoids standing in racks, waiting. Here's what that number means. One plant, 10,000 units a year, opened this month. Add Agility, Figure, Unitree, Tesla and AgiBot, and humanoids stop being scarce fast. When supply stops being the problem, the story becomes what happens to the units that get replaced. A Walker S built this year still works perfectly in 2029, whether or not it's the newest thing on the floor. That's what robotautotrader.com is for. Mass production creates the first real generation of used robots, and somebody has to make them findable.

This is the strongest angle: a £999 flagship dropping to £299 in under two years. The Dyson 360 Vis Nav launched at $999. It's now $299.99 on Woot, a 70% cut in under 24 months. That's not a fire sale, that's the normal robot vacuum depreciation curve, and it's the number every used-market seller needs to know before they price their own unit. Here's the bit that matters for owners: independent testing still shows 89%+ pickup on debris and a proper HEPA seal at 99.99%, so the cleaning hardware hasn't aged. What's dragging the resale value down is software: recurring Wi-Fi drop-outs and inconsistent mapping that Dyson hasn't fully fixed since launch. Buyers are pricing in the firmware risk, not the suction. If you're selling a used 360 Vis Nav: don't price off the original RRP, price off what a fixed-navigation robot is actually worth to a buyer today, closer to $250 to $300 territory even in good condition. If you're buying: at $299 the hardware case is strong, but ask the seller to demo a full unattended run on your Wi-Fi network before you hand over cash, connectivity issues don't show up in a five-minute showroom test. Is unreliable navigation software now a bigger driver of robot vacuum resale value than the suction spec ever was?

Seventeen humanoids in one frame. Look at the two ends of that lineup. 👀 Figure 03 shows up twice- same body, once with Helix & once with Helix 2.5. Same hardware, newer brain, listed as a different robot. And that second one is a day old. That's the whole industry in one image. The bodies are converging: two arms, two legs, roughly human height, black or white shell. What changes is the software on top. Which means depreciation here looks nothing like cars. No engine, no rust, no mileage. A frame built last year still lifts what it lifted, and the intelligence arrives by update. Seventeen models now. Next year there will be thirty, and plenty of these will be sitting in storage rooms working perfectly. That's exactly what robotautotrader.com is built for: listing the units that got replaced instead of used up. Sell the one you've outgrown, buy the frame someone else already paid retail for, and let the software catch it up. The upgrade cycle creates the inventory. We just make it findable.

A Unitree G1 in a blue clown wig threw a scripted roundhouse kick at a tourist demo in Xinjiang and caught a boy in the stomach. He wasn't seriously hurt. The clip has 23,000+ posts on it because everyone can see how close it came to being worse: this is a 127cm, 35kg robot with real servo torque, not a toy. Here's the bit that matters if you're actually shopping for one of these: the G1 is a commercial product, not a lab prototype. Universities, robotics clubs and service companies buy them right now, list price starts around $16,000 for the base config. There's no bystander safety certification for humanoid robots doing pre-programmed martial arts routines near the public the way there is for industrial cobots on a factory floor with light curtains and force limits. If you're buying or hiring one out for events, the checklist is simple: confirm there's a physical e-stop within reach of the operator, ask what the torque limit is in "performance" mode versus "safe" mode, and get it in writing who's liable if the choreography goes wrong near a crowd. Should humanoid robots need a certified safety mode before they're allowed to run scripted routines near untrained bystanders, same as cobots already do on factory floors?

Odyssey just showed one AI model controlling robots, humanoids, cars & drones. Same brain, different bodies. That's bigger news for used robotics than anything else so far this year. Right now a robot's value is bundled- it’s hardware plus whatever intelligence shipped inside it. Split those apart & the body is just actuators, sensors & a frame. Software depreciates. Steel doesn't. Buy the body secondhand at robotautotrader.com. Let the intelligence arrive by update.