基于自学习中枢模式发生器的仿人机器人适应性行走控制

为了克服传统中枢模式发生器(Central pattern generator, CPG)关节空间控制方法的复杂性和局限性, 本文基于自学习中枢模式发生器模型, 提出了一套在线调制和融合多传感器信息的仿人机器人环境自适应行走控制方法.算法难点在于如何在机器人的工作空间将自学习CPG用于工作空间轨迹生成, 并使CPG参数直接和步态模式相关联.本文提出了利用自学习CPG来学习和实时生成机器人质心轨迹和脚掌轨迹的方法, 在线调节机器人步长、抬腿高度和步行速度等关键参数.参考生物反射行为, 利用传感反馈信息激发CPG以产生具有环境适应性的工作空间轨迹, 提升行走质量. 控制系统的参数通过优化算法来进一步改善行走性能.相比于传统的CPG关节空间法, 本文所采用的自学习CPG工作空间法不仅极大简化了CPG网络结构而且提高了仿人机器人行走的适应性.最后, 通过仿人机器人坡面适应性行走的仿真和实验, 验证了所提出控制策略的可行性和有效性.

Adaptive Locomotion Control of Humanoid Robot Based on Self-Learning CPG

To overcome the complexities and limitations of the joint-space CPG (central pattern generator)-inspired control methods, a novel CPG inspired workspace control strategy is presented in this work, where humanoid adaptive workspace trajectories is generated online through the self-learning CPG (SL-CPG). The challenge of this work is that how to generate the workspace trajectory of the humanoid robot in the workspace via SL-CPG and how to directly connect the SL-CPG parameters with the walking gait mode. In this paper, a novel method to generate the foot trajectory and center of mass (CoM) trajectory by using two sets of SL-SPG is proposed, and in this way some key parameters such as robot step size, leg height, and walking speed can be adjusted online. The resultant workspace trajectories can be adjusted by the sensory information, which mimics the vestibular reflex of animals. Furthermore, an evolutionary algorithm is developed to tune the control system parameters to improve the walking performance. Compared with the traditional CPG joint-space methods, the SL-CPG workspace method adopted in this paper not only greatly simplifies the CPG network structure but also improves the adaptability of humanoid robot walking. Finally, the applicability of the proposed control strategy is demonstrated through simulations and experiments focusing on the humanoid robots gait pattern adaptation over sloped terrain.