RLHarness: Co-evolving Procedural Skills with Reinforcement Learning for Long-horizon Multimodal Reasoning
RLHARNESS is introduced, which organizes Skills, selection and execution protocols, few-shot demonstrations, and task contracts into a unified, versioned Harness and alternates Harness evolution with policy learning, showing that an evolving Harness complements RL by continually updating the external program that the p...
Zi-Qiao Shang, Zian Xu, Ji-Chen Yan et al.
· 0 citations