Evaluating an agent's planning ability remains challenging for open-domain tasks, where verifying procedures through execution is often impractical. Existing…
机构:UCL
来源:arXiv 2610.26679 | AI4Papers 论文推荐平台