弹性支撑条件下的灵长类仿生机器人悬臂运动控制方法研究

In gibbon brachiation mechanism and primate bionic robot researches, the supporting bars are typically assumed to be rigid, which ignores the elasticity of the supporting branches in real gibbon brachiation motion. This project targets a two link primate bionic robot and studies its brachiation motion control problem under elastic support condition. Firstly, based on the Euler–Bernoulli beam model and the assumed mode method and dual spring-dampers oscillator model, the dynamic model of primate bionic robot under elastic support condition is derived. Then according to its high dimensional rigid-flexible hybrid underactuated characteristics, a systematic brachiation motion planning and Lyapunov based trajectory tracking method are proposed by extending the virtual constraint method. For the time-varying problem of the brachiation target brought by the vibration of the elastic supporting bar, a target redirecting method is proposed to realize the dynamic target tracking control. Finally, the brachiation energy model is build by synthesizing the brachation motion planning and control. Based on the energy model, the brachiation locomotion mechanism and energy efficient brachiation trajectory planning problems are solved. This project aims to explore the gibbon’s origin of the ability of adapting the natural environment and enhance the capability of the primate bionic robot’s adaptability in unstructured complex environment. Furthermore, control research of high-dimensional rigid-flexible hybrid underactuated mechanical systems also has important scientific significance.

长臂猿悬臂运动机理研究和灵长类仿生机器人研究中普遍将支撑杆假设为刚性，忽略了长臂猿真实运动中支撑树枝的弹性。本项目以双臂式灵长类仿生机器人为对象，研究其在弹性支撑下的悬臂运动控制问题。首先以Euler-Bernoulli梁模型结合假设模态法和双弹簧阻尼器振子模型，建立弹性支撑下灵长类仿生机器人动力学模型；然后针对其中的高维刚柔混合欠驱动特性，对虚约束方法进行推广，提出一种系统化的悬臂运动轨迹规划和基于Lyapunov的轨迹跟踪控制方法；同时针对弹性支撑杆振动带来的悬臂运动目标的时变问题，提出一种目标重定向方法来实现动态目标跟踪控制；通过综合悬臂运动轨迹规划与控制，建立悬臂运动能量模型，并以此解决弹性支撑下悬臂运动机理和高能效悬臂运动优化等问题。本项目将探寻长臂猿适应自然环境能力的根源，增强灵长类仿生机器人在复杂环境中的适应能力。另外，对高维刚柔混合欠驱动机械系统的研究也具有重要的科学意义。

灵长类动物具有类人的身体结构以及以悬臂形式为核心的运动特征。研究灵长类仿生机器人对野外机动能力提升具有很重要作用。现有研究普遍将支撑杆假设为刚性，忽略了长臂猿真实运动中支撑树枝的弹性，本项目以双臂式灵长类仿生机器人为对象，研究其在弹性支撑下的悬臂运动控制问题。首先以Euler-Bernoulli梁模型结合假设模态法和双弹簧阻尼器振子模型，建立弹性支撑下灵长类仿生机器人动力学模型；然后针对其中的高维刚柔混合欠驱动特性进行研究与分析，证明弹性假设下系统可控性降低，存在不控子系统；结合遗传算法，研究不同约束条件下悬臂运动优化方法；利用深度强化学习方法对悬臂运动策略进行学习，分别从网络结构、先验知识编码和演示数据融合等角度对强化学习方法进行改进，以解决其奖励稀疏、训练过程缓慢、需要大量环境交互、运动不平滑等问题；建立了一套系统的悬臂飞跃运动规划与控制方法，将其划分为三阶段，推导起飞、飞行和降落阶段的边界条件，融合环境信息和人为指定条件，解决普遍意义下的悬臂飞跃控制问题。搭建了电机驱动和气动肌肉驱动的双臂式灵长类仿生机器人系统和刚性/弹性支撑悬臂运动实验平台。开展实验研究，结果表明，所提出的方法可以有效实现灵长类仿生机器人悬臂运动。