Axis Robotics

What Is Axis Robotics (AXIS)? Launched on January 19, 2026 by a team based in Singapore, Axis Robotics is a data platform for Physical AI that collects, verifies and processes training data for robots. Robot learning models need large, varied datasets of physical interaction, and collecting that data with real robots in laboratories is slow and expensive. Axis gathers it from a global pool of contributors instead. The platform draws on two sources. The first is simulation: contributors use a desktop browser to teleoperate a simulated robot arm through manipulation tasks. The simulator, MuJoCo compiled to WebAssembly, runs locally in the browser, so contributors do not need robot hardware. The second source is ego-centric human data, meaning first-person recordings of real-world activity captured with mobile devices and specialized equipment. A mobile app for this is in development. Each task moves from pre-training through training to post-training. In pre-training, contributors record demonstrations that train a base policy. In post-training, that policy tries the task on its own while a contributor watches, and the contributor takes over when the policy fails. These corrections train the next version of the policy, and whatever the policy still gets wrong decides which tasks are published next. Axis calls this loop a "compounding data engine". Every submitted trajectory is replayed in simulation to verify it, then scored, cleaned and expanded into domain-randomized training samples by an augmentation pipeline built on IsaacSim. The team trains robot policies and world models on this data. In a paper on arXiv (2607.21588), further pre-training Physical Intelligence's π0.5 model on Axis data raised its LIBERO-Plus success rate from 83.9% to 88.8%, 31.3 points above a control trained on the same amount of RoboCasa365 data. Who Is Behind Axis Robotics? The team includes AI and robotics researchers from UC Berkeley, Carnegie Mellon University, UCLA and Nanyang Technological University, along with members who have previously built consumer products used by more than 30 million people. Organizations working with Axis include Booster Robotics, Manycore, Geely Lotus and Dexmal AI. What Makes Axis Robotics Unique? Axis records where each piece of data came from on Base. When a contributor signs a verified trajectory, it gets a Data ID, which is written on-chain together with its task ID and contributor ID. Contributors sign in with email or X, and the app provides a custodial wallet. A signature is needed only to record a contribution on-chain and make it eligible for rewards. Quality checks are fully algorithmic: replay verification and quality thresholds, with no manual review. As of October 8, 2026, the platform had more than 224,000 users, 7.2 million collected trajectories and 6.8 million on-chain transactions. How Many AXIS Tokens Are There in Circulation? AXIS has a fixed total and maximum supply of 1,000,000,000 tokens. All tokens are allocated at genesis. There is no inflation and no way to mint more. At the token generation event (TGE), 165,000,000 AXIS (16.5% of supply) will circulate: 80,000,000 from the Ecosystem allocation, 60,000,000 for liquidity, 20,000,000 from Contributor Rewards, and 2,500,000 each for Sonar participants and the Kaito Content Creator Reward Pool. The remaining tokens unlock on different schedules, and all allocations are fully unlocked 35 months after the TGE. The token has two proposed uses: contributors earn AXIS rewards from Program budgets based on verified work quality, and approved sponsors use AXIS to fund data-collection Programs. How Is the Axis Network Secured? AXIS is issued on Base, an Ethereum layer-2 network. The token and the on-chain provenance records rely on Base's security. Axis also replays every trajectory in simulation before scoring it, so submissions that cannot be reproduced do not count.