轮腿混合式机器人力学变分原理与协调运动控制

The rapid response of disaster scene requires mobile robots to be adaptable to a wide range of terrain, so that it can not only climb and crawl on the ruins, but also move quickly and efficiently on the flat road. This project combines model-driven and data-driven methods to study the coordinated motion control of hybrid wheel-legged robots in a wide range of terrain, including:(1) analyzing the ‘leg-wheel-terrain’ interaction mechanism, obtaining the motion characteristics of hybrid wheel-legged robots under various motion modes and establishing its online locomotion modes decision-making neural network; (2) studying the mechanical variational principle and the structure-preserving numerical discretization methods of the hybrid wheel-legged mechanism under multiple constraints/external forces, establishing the accurate and credible discrete dynamic model of the system; (3) establishing the control policy learning architecture of the hybrid wheel-legged robots in numerical simulation environment, developing efficient policy optimization algorithms under the proposed architecture and transferring the learned control policy to the real physic system. The purpose of this project is to provide theoretical basis and technical support for the safe, stable and efficient search and rescue operation of hybrid wheel-legged robots in complex and polytropic terrain by fully combining the speed and efficiency of the wheel mechanism with the dexterous adaptability of the leg mechanism.

灾难现场的快速响应，需要移动机器人具有广域地形环境的适应能力，使其既能在废墟上攀爬行走，又能在平整路面快速、高效移动。本项目结合模型驱动和数据驱动的方法研究广域地形环境中轮腿混合式机器人协调运动控制问题，具体包括：（1）分析广域地形‘腿-轮-地’交互作用机理，得到轮腿混合式机器人各类运动模式下的运动特性，建立其运动模式在线决策神经网络；（2）研究多约束/外力作用下的多自由度轮腿混合机构力学变分原理和其动力学模型的保结构数值离散方法，建立系统精确、可信的离散动力学模型；（3）搭建数值仿真环境中轮腿混合式机器人控制策略学习框架，开发该框架下高效的策略优化算法，实现所学控制策略在实际物理系统中的迁移应用。本项目旨在充分结合轮式机构的快速、高效性和腿式机构的灵巧适应能力，为实现轮腿混合式机器人在复杂多变地形环境中安全、稳定、高效搜救作业提供理论基础和技术支持。

为实现轮腿混合式机器人在复杂地形环境中的高性能运动，项目系统研究了轮腿混合式机器人多运动模式下的动力学建模与仿真、多传感器融合状态估计与环境感知、全局路径规划与局部轨迹优化、动力学参数辨识与控制。通过项目研究，建立了基于Lagrange-d’Alembert原理的离散动力学模型，开发了基于视觉惯性里程计的鲁棒定位与三维地形重建算法以及多种运动模式下的高效率路径规划与轨迹优化算法，搭建了轮腿混合式机器人软、硬件平台，实现了轮腿混合式机器人基于高斯过程回归的整机参数辨识与高性能模型预测控制方法。该研究使轮腿混合式机器人能充分结合轮式机构的快速、高效性和腿式机构的灵巧适应能力，最终实现其在复杂多变地形环境中的安全、稳定、高效作业。