两足机器人步态综合研究进展

Review of Studies on Biped Robot Gait

把两足机器人步态综合方法分为参考轨迹法和自然动力学法两大技术流派.根据参考轨迹获得方法的不同,参考轨迹法又分为步行数据法、中枢模式发生器法和动力学模型法.自然动力学法也可分为被动动力学法和虚拟模型法.在总结两大流派研究进展的基础上,分析了它们各自的优势和不足,指出了今后工作的重点,即参考轨迹法需要提高步行效率和轨迹的生成速度,而自然动力学法需要增加功能.最后分析了强制学习技术在步态综合中应用的适应性和多自由度造成的组合爆炸问题.

Gait designs were divided into two main categories： reference trajectory and natural dynamics. The former is based on walking data, central pattern generator or dynamic model. The latter is classified into passive dynamics and virtual model control. The advantages and disadvantages of the two main gait designs were analyzed. The future study is to increase walking efficiency for reference trajectory, and to add more functions for natural dynamics. The application of reinforcement learning in gait design and the problem of the curse of dimension were discussed.