Trajectory2Gym：面向工具使用智能体的决策接地组合交互合成
原标题：Paper page - Trajectory2Gym: Decision-Grounded Compositional Interaction Synthesis for Tool-Use Agents
AI 摘要
该论文提出 Trajectory2Gym 框架，用于训练工具使用智能体。它不再单纯收集成功轨迹，而是围绕覆盖不足的组合结构构造可执行交互，并只从可见历史支持的决策中提取监督信号。在五个 Qwen3 规模上，该方法在 τ-bench、τ²-bench 和 BFCL 上均取得提升，三项基准等权平均提高 16.9 至 29.8 个百分点，多轮有状态交互中增益最明显。
正文节选
Trajectory2Gym: Decision-Grounded Compositional Interaction Synthesis for Tool-Use Agents Abstract Trajectory2Gym improves tool-use training by constructing interactions with compositional coverage and decision grounding rather than collecting successful trajectories. Successful tool-use trajectories are not necessarily useful supervision. Even successful interactions may underrepresent multi-step dependencies or contain actions that are unjustified by the information visible to the agent at dec