仿人机器人高能效自激步态生成与平衡控制方法研究

Improving energy efficiency of humanoid robots is becoming significant for humanoid’s application. Investigations unveil that human’s walking gaits are generated by the spinal cord, and the adjustment of the brain and muscles are the key issue for energy efficiency and balancing ability. Up to now, humanoid self-excitation gait does not have the feature of high energy efficient, because it only simulates the spinal cord but not the brain and muscles. The study of humanoid self-excitation gaits based on the adjustment mechanism of the brain and muscles is a basic scientific problem needed to be solved...In this project, a self-excitation gait generation and control model of humanoid robot will be built, and the intrinsic relationship between the input and output of the generator will be found. Methods of energy-efficient self-excitation gait generation based on reinforcement learning and the reflective balance control with anthropomorphic muscle viscoelasticity will be studied with the adjustment mechanism of the brain and muscles. Mapping disciplinarian of parameters of the self-excitation gait generator, the coordinated gait which matches the physical characteristics of the robot and the balance control based on interaction force coupling optimization will be solved. This project will provide a new method for humanoid walking with high energy efficiency, which have important academic significance and practical value.

提高仿人机器人的行走能效是延长持续运行时间的关键，对其应用具有重要意义。研究表明，人体步态由脊髓自激产生，而人脑和肌肉的调节作用对步态的能效、平衡性起着决定性作用。目前，仿人机器人自激步态无高能效的特性，其原因在于仅模拟了脊髓产生步态的机理，尚未借鉴人脑和肌肉对步态的协调、平衡控制机制。研究借鉴人脑和肌肉调节机制的仿人机器人自激步态生成与控制方法，是亟需解决的基础科学问题。..本项目将仿生学与工学相结合，借鉴人脑和肌肉对自激步态的调节机理，建立仿人机器人高能效自激步态控制器模型，探索自激步态生成器的输入输出参数间的内在联系规律，研究基于强化学习的高能效自激步态生成方法、具有拟人肌肉粘弹性的反射平衡控制方法等，解决自激步态生成器的参数映射规律、匹配机器人物理特性的协调步态生成方法、作用力耦合优化的平衡控制等科学问题，为仿人机器人高能效稳定行走提供一种新方法，具有重要的学术意义及实用价值。

仿人机器人是一个多关节、非线性、强耦合的复杂系统，仿人机器人通过关节协调耦合运动传递身体动能和势能，但主动驱动行走能效低的问题日益凸显。.本项目围绕仿人机器人高能效自激步态生成与平衡控制前沿理论与方法，建立了基于中枢神经反射原理的双中央模式发生器（CPG：Central Pattern Generators）模型，并在多组神经元之间进行相互刺激或抑制，通过调整振荡器参数，获得了输入与输出的内在映射规律，实现行走动作的自校正；提出了基于空间量化动力学理论的直膝行走步态生成方法，能效较屈膝步态行走时提升67%；在行走过程中为快速应对冲击型扰动的问题，利用监督学习对基于李群的非线性卡尔曼滤波进行精确状态估计结果修正的基础上，提出了基于质心加速度变化的三步落脚点调整算法，并借鉴了人体的肌肉耦合机理，构建了仿人机器人拟人肌肉粘弹性模型的柔顺控制器来减少机体振动，使得机器人具备了高能效自激步态调整和快速恢复稳定行走能力。.本项目提出的方法在已研制的BHR-6P仿人机器人上进行了验证，实验结果表明项目所提出的技术方案提高了仿人机器人高能效运动和平衡控制的能力，推进了仿人机器人走向实际应用。