French startup solves Rubik’s cube; modular hand launches

Robot hands are having a moment. Not the old kind — pinch a block, lift a cup, repeat — but the much harder kind where fingers have to coordinate, regrasp, feel contact, and keep an object stable while it moves. That is why a Rubik’s Cube still gets attention. It is not just a toy. It is a stress test for in-hand manipulation. This week, two different moves pushed that story forward: Genesis AI showed a humanoid hand solving one, and DexRobot’s modular DexHand021 kept gaining traction as a developer-ready hand teams can actually integrate. (genesis.ai) ### Why does a Rubik’s Cube still matter? A cube solve forces a robot hand to do several hard things at once — hold the object, rotate one face without losing the whole cube, shift finger positions mid-task, and recover from tiny errors before they snowball. Robots have solved cubes before, including OpenAI’s 2019 Dactyl work, but that was exactly the point: the benchmark is old, while reliable humanoid-hand dexterity is still not solved at product level. (arxiv.org) ### What did Genesis AI actually show? Genesis AI launched GENE-26.5 on May 6 and framed it as a general manipulation model, not a one-off cube trick. In its own materials, the company showed the same stack handling cooking, smoothie making, wire harnessing, lab automation, multi-object grasping, piano playing, and Rubik’s Cube solving with a human-scale robotic hand. Reuters also tied the launch to a broader fu(arxiv.org) backed by Eric Schmidt and Xavier Niel. (genesis.ai) ### Why pair the brain with the hand? Because dexterity breaks if the hardware and training setup fight each other. Genesis is basically arguing that the “brain” and the hand have to be co-designed — model, simulator, data collection gloves, and physical hand all in one loop. That matters because contact-rich manipulation is messy. A finger lands a millimeter off, friction changes, the object shifts, and the policy has to adapt in real time. (genesis.ai) ### So where does the modular hand fit in? This is the less flashy but maybe more important part. DexRobot’s DexHand021 is not a viral demo first. It is a product hand for labs and integrators. The hand has 19 degrees of freedom, with 12 active and 7 passive, plus sensing for position, force, slip, and proximity. Its pitch is simple: modular parts, ROS1 and ROS(genesis.ai)instead of fabricating a hand from scratch. (humanoid.guide) ### Why is “modular” such a big deal? Because robot hands break, get tuned, and get swapped constantly. A modular hand is like a dev board for manipulation — less magical than a polished humanoid demo, but much more useful if you are training policies, testing grasp planners, or integrating with an arm. Quick replacement and standard software hooks cut the cost of experimentation. That is what tur(humanoid.guide)w. (humanoid.guide) ### Is this just another demo cycle? Partly, yes. Public dexterity showcases can overstate robustness. A filmed solve does not mean all-day warehouse reliability. But the mix has changed. The interesting thing is not just “look, fingers moved nicely.” It is that the showcase side and the developer-tool side are starting to meet — one company selling the dream of general manipulation, another selling the hand, SDK, and simulation stack needed to chase it. (genesis.ai) ### What should people watch next? Watch for repeatability, not spectacle. Can these hands recover from slips? Can outside teams use the SDKs and get decent results? Can one hand design support multiple tasks without a ton of per-task tuning? Those are the questions that separate a cool cube solve from a real platform. (genesis.ai)robot solved a Rubik’s Cube. That part is old. The news is that dexterous hands are starting to show up as both public proof and purchasable infrastructure — and that is how a benchmark turns into a market. (arxiv.org)