RoboACE: Adversarial Co-Evolution for Robot Self-Improvement
arXiv:2611.13034v1 Announce Type: cross Abstract: Robot agents can improve reusable skills, planning procedures, and recovery strategies from execution feedback, yet a fixed practice source leaves the search for new weaknesses unchanged. Useful practice depends on the agent's current capabilities, so the procedure that chooses practice must also be revisable. We present RoboACE (Robot Adversarial Co-Evolution), a framework that jointly evolves curriculum and execution harnesses around frozen fou
阅读原始来源来源记录
项目分析
暂无五维分析,可先查看上方来源资料。