For humanoid robots to live up to their potential utility, they must be able to robustly recover from instabilities. In this work, we propose a number of balance enhancements to enable the robot to both achieve specific, desired footholds in the world and adjusting the step positions and times as necessary while leveraging ankle and hip. This includes improving the calculation of capture regions for bipedal locomotion to better consider how step constraints affect the ability to recover. We then explore a new strategy for performing cross-over steps to maintain stability, which greatly enhances the variety of tracking error from which the robot may recover. Our last contribution is a strategy for time adaptation during the transfer phase for recovery. We then present these results on our humanoid robot, Nadia, in both simulation and hardware, showing the robot walking over rough terrain, recovering from external disturbances, and taking cross-over steps to maintain balance.

翻译：为使仿人机器人充分发挥其潜在效用，其必须具备从失稳状态中稳健恢复的能力。本文提出多项平衡增强技术，使机器人既能在环境中实现特定期望的落脚点，又能利用踝关节与髋关节灵活调整步幅位置与着地时机。这包括改进双足运动捕获区域的计算方法，以更全面考虑步态约束对恢复能力的影响。随后我们探索了一种基于交叉步维持稳定性的新型策略，该策略显著扩展了机器人可恢复的跟踪误差类型。最后一项贡献是提出恢复过程中转移阶段的时间自适应策略。我们在仿人机器人Nadia上分别通过仿真与实物实验验证了上述成果，展示了机器人跨崎岖地形行走、从外部扰动中恢复以及通过交叉步维持平衡的能力。