Browser robot training on‑chain

Robot training is moving into the browser: Axis Robotics says anyone with a desktop browser can control a simulated robot arm online and upload the resulting training record to the Base blockchain. (docs.axisrobotics.ai) The basic idea is imitation learning, a method where robots learn from examples instead of handwritten rules. Axis says each browser session creates a “trajectory,” a timestamped log of robot joint states, object positions, control inputs, and task metadata. (axisrobotics.ai) (docs.axisrobotics.ai) Axis’s Hub is live at hub.axisrobotics.ai, where recent site data showed 403,499 data trajectories, 27,787 contributors, and 303,016 transactions. Task cards on the site include simulated chores such as lining up bottles, hanging a hanger, and watering a flower. (hub.axisrobotics.ai) To contribute, users sign in through Privy with email, Google, X, or an existing wallet, then start a task in Chrome, Firefox, Safari, or Edge on a desktop or laptop. Axis says the simulation runs locally in the browser through WebAssembly, with no separate download and no graphics processing unit required. (docs.axisrobotics.ai 1) (docs.axisrobotics.ai 2) In plain terms, WebAssembly is a way to run near-native code inside a web page, and blockchain storage here is being used as a public receipt rather than a place to run the robot. Axis says accepted trajectories receive a unique Data ID on Base, Coinbase’s layer-2 network, to create a provenance and ownership trail. (docs.axisrobotics.ai 1) (docs.axisrobotics.ai 2) The company’s technical documents say the browser is only the first step in the pipeline. After upload, trajectories are cleaned, smoothed, and replayed on graphics processor-backed Isaac Sim servers, where the same action can be regenerated under varied lighting, textures, and scene conditions to make the training set larger. (techreport.axisrobotics.ai) (docs.axisrobotics.ai) That workflow targets a central robotics problem: collecting enough varied examples is expensive when every attempt needs a real machine, a safe room, and human operators. Axis argues browser teleoperation lowers the cost of data collection, while higher-fidelity replay on server hardware adds realism later in the pipeline. (techreport.axisrobotics.ai) (axisrobotics.ai) Axis said on March 26, 2026 that it was moving from testing to broader availability after two community test rounds that generated nearly 300,000 robotic trajectories. The company’s product guide also says reward details in United States dollar coin and token form are still under development, while peer review payments for rating other users’ runs are planned. (accessnewswire.com) (docs.axisrobotics.ai) Axis is pitching that mix of browser access, shared data collection, and on-chain recordkeeping as a way to let far-flung contributors work on the same robot models without sharing a lab. The test for that claim will be whether those browser-made trajectories produce policies that still work when the robot leaves the web page and touches the real world. (docs.axisrobotics.ai) (techreport.axisrobotics.ai)