HARBOR trains robot locomotion policies from a single prompt, fully autonomously in 1.5 hours

breadli428 · x · 2026-09-05

HARBOR autonomously handles the entire robot policy training loop: from one prompt, it builds the task, designs rewards, trains, tunes, and evaluates a locomotion policy in simulation — end to end in about 1.5 hours. The work has been accepted at CoRL 2026, showing how far agent-driven RL training has come.

Original post →

More from coding & agent

coding & agent channel →