
OpenRoboto is a live robotics-model competition in which miners fine-tune a π0.5 vision-language-action model and validators rank reproducible simulation evaluations.
Miners fine-tune the open π0.5 VLA base, merge any adapter into a complete checkpoint, upload that checkpoint to Hugging Face, burn the submission fee and commit the exact artifact on-chain. Backend workers evaluate submissions on randomized LIBERO tasks in MuJoCo, and validators read the settled ranking and submit weights.
The public API, current round, leaderboard and evaluation queue are live, and submitted model checkpoints are public artifacts.
Robot-learning researchers and Bittensor miners with GPU and Hugging Face infrastructure who can train and submit vision-language-action checkpoints.
There is no published end-user price; miner evaluation submissions require a 0.1 TAO burn, which is destroyed rather than paid to the operator.
Every rewarded submission is a complete open robotics checkpoint evaluated with randomized, reproducible LIBERO/MuJoCo scores rather than a closed hosted robot-control service.