📄 中文摘要
人形机器人在现实环境中需要执行灵活且富有表现力的全身运动。现有的文本到运动生成模型主要基于捕获的人类运动数据集进行训练，这些数据集的先验假设基于人类的生物力学、驱动方式、质量分布和接触策略。当这些运动直接转移到人形机器人时，生成的轨迹可能满足几何约束（如关节限制和姿态连续性），并在运动学上看起来合理。然而，这些运动往往违反了现实执行所需的物理可行性。为了解决这些问题，提出了PhyGile，一个统一框架，旨在闭合机器人原生运动生成与通用运动跟踪（GMT）之间的循环。
PhyGile: 基于物理前缀的灵活运动生成用于敏捷的通用人形运动跟踪
📄 English Summary
PhyGile: Physics-Prefix Guided Motion Generation for Agile General Humanoid Motion Tracking
Humanoid robots are required to perform agile and expressive whole-body motions in real-world environments. Existing text-to-motion generation models are primarily trained on captured human motion datasets, which assume human biomechanics, actuation, mass distribution, and contact strategies. When such motions are directly retargeted to humanoid robots, the resulting trajectories may satisfy geometric constraints (e.g., joint limits and pose continuity) and appear kinematically reasonable. However, they often violate the physical feasibility necessary for real-world execution. To address these challenges, PhyGile is proposed as a unified framework that closes the loop between robot-native motion generation and General Motion Tracking (GMT).