droid8 results

π0.5 (openpi JAX) trained with FedAvg / centralised on 8 DROID buildings, evaluated in RoboLab and on a real DROID robot. Sign in, upload results, logs and videos: each submission becomes a pull request on DecentVLA/droid8-results, made with your own sign-in, which the maintainer merges.

Sign out

Upload a run

Merged runs Refresh

A whole folder from the command line (also opens a pull request, no write access needed):
hf upload DecentVLA/droid8-results ./my_results runs/<date>_<you>_<arm>_<eval>_<name> --repo-type dataset --create-pr