📄 中文摘要
在运动控制任务中，深度强化学习（DRL）已展现出高性能，但学习到的策略的决策过程仍然是一个黑箱，使得人类难以理解。已知在步态等周期性运动中，存在隐含的运动阶段，如支撑阶段和摆动阶段。基于此，研究假设为运动控制训练的策略可能也代表一种可被人类理解的阶段结构。为验证这一假设，研究考虑了一个适合观察策略是否通过与环境的互动自主获取时间结构化阶段的运动任务。
揭示运动策略中的潜在阶段结构和分支逻辑：以半猎豹为例
📄 English Summary
Uncovering Latent Phase Structures and Branching Logic in Locomotion Policies: A Case Study on HalfCheetah
In locomotion control tasks, Deep Reinforcement Learning (DRL) has shown high performance; however, the decision-making process of the learned policy remains a black box, making it difficult for humans to interpret. It is well established that implicit motion phases exist in periodic movements such as walking, including the stance phase and the swing phase. This study hypothesizes that a policy trained for locomotion control may also embody a phase structure that is interpretable by humans. To test this hypothesis in a controlled environment, a locomotion task is considered that allows for the observation of whether a policy autonomously acquires temporally structured phases through interaction with the environment.