1 paper · 1 filter
Lin Wu, Shu Yao, Yuheng Wang +11
Training multi-agent collaboration requires trajectories of checking, feedback, and repair, yet task-solving rollouts may end in immediate approval or repeat unsuccessful attempts…