AgenticRoboRL: Bringing Coding-Agent Post-Training to Agentic Robotics
A 9B-parameter robot planner’s success rate rose from 18.1% to 71.0% on LIBERO-PRO without external memory.
AgenticRoboRL combines supervised fine-tuning with multi-turn reinforcement learning to improve the planner itself. Harness-specific adapters link planner interactions to task outcomes, supporting both VLA skill invocation and control-code generation while keeping execution components fixed. The authors say they will release the framework and trained planner checkpoints.
HF Daily Papers' note
AgenticRoboRL combines supervised fine-tuning with multi-turn reinforcement learning to improve the planner itself. Harness-specific adapters link planner interactions to task outcomes, supporting both VLA skill invocation and control-code generation while keeping execution components fixed. The authors say they will release the framework and trained planner checkpoints.
HF Daily Papers' note
score 5