EvoHIL: Self-Evolving Reward and Flow-Matched Policy Optimization for Robust Human-in-the-Loop Reinforcement Learning

EvoHIL is a human-in-the-loop reinforcement learning framework for contact-rich robotic manipulation. It combines a self-evolving success classifier, flow-matched generation of temporally coherent action chunks, and retention-aware offline fine-tuning with relit interaction data. The abstract reports experiments on six tasks using Franka FR3 and SO-101 arms under a controlled lighting shift, claiming improvements in success, agreement with human-confirmed labels, motion smoothness, and completion time. However, the supplied abstract provides no quantitative results, and the listed August 2026 publication date is future-dated and cannot currently be verified.