Hugging Face · 每日论文·2026-10-13 04:00· 2 天前AI 评分46

WorkRover: 通过递归环境演化扩展长时地平线协作代理训练

WorkRover: Scaling Long-Horizon Cowork Agent Training Via Recursive Environment Evolution

AI 导读

WorkRover 框架通过递归环境演化实现长时地平线协作代理的训练，其核心机制包括 Factorial Compatibility Replay (FCR) 和 Verifier-to-Turn Credit Assignment (VTCA)。