Recent work shows that a frozen robot policy performs far better inside an agent harness, where control code decides when to query the policy, consult vision…
机构:Stanford
来源:arXiv 2610.42141 | AI4Papers 论文推荐平台