arXiv · cs.AI · Yuanhao Li, Hongbo Wang, Xuhong Chen, Yiming Cao, Xunzhu Tang · 10 小时前模型评分46
反事实轨迹重放（CRR）：用可分叉环境为软件工程智能体提供免费过程奖励
Counterfactual Rollout Replay: Forkable Environments as Free Process Rewards for Software Engineering Agents
导读
研究者提出反事实轨迹重放（CRR，Counterfactual Rollout Replay），利用可分叉的可执行环境在选定决策点恢复状态、采样替代动作并向前推演，用真实轨迹终点回报与反事实回报之差替换该步优势值，无需人工过程标注或过程奖励模型。
分类置信度
中 · 待人工复核
- 只有单一媒体来源，需要人工复核。
来源级别：媒体与个人
事件类型：研究发布
证据摘录
- “We introduce Counterfactual Rollout Replay (CRR), a training-time procedure that uses forkable executable environments to obtain step-level return contrasts.”
- “The method needs no human process labels or learned process reward model; free refers to those supervision costs, not replay compute.”
- “These results apply to environments with affordable, reliable state restoration; stochastic continuations and expensive or imperfect replay remain limitations.”
来源：arXiv · cs.AI · arxiv.org