基于参数最优的仿人机器人复杂运动规划研究

Optimal trajectory planning of joints is the key technology to achieve complex motions for humanoid robot in different environment. This project focus on humanoid robot's soft-start and soft-brake for fast walking, falling motion with external force while being unable to effectively brake, and states switch problem between different movements of humanoid robot to achieve the optimal control of continuous movements for humanoid robot in different environments. Firstly, theories of soft-start and soft-brake for humanoid robot are analyzed deeply. An algorithm based on parameters optimum and enhanced control is present in an innovative way. The parameterized controllers are constructed by number of piecewise constant functions to transfer the optimal control problem into parameters optimum problem. Secondly, a staged control method of falling motion for humanoid robot with external force is proposed. An adaptive algorithm of falling control is designed to achieve optimum in threes aspects: the ground impact in touchdown moment, the location of the touchdown and the stability after falling. Meanwhile, An constraint transformation technique is adopted to smooth the switching between fast walking and falling motion, and the constraints of the state function at the origin are smoothing. An imporved operator and an initial state screening set are used to correct the feasible direction automatically and accelerate the optimization of processing. Finally, experiments on continuous movement between fast walking and falling is achieved on humanoid robot to verify the reliability and validity of the proposed control method.

关节最优轨迹规划是仿人机器人实现各种环境下复杂运动的关键。本项目研究仿人机器人快速步行运动的软启动、软制动以及外力干扰、无法有效制动情况下的倒地柔顺控制和不同运动状态之间的切换问题，实现多种环境下仿人机器人连续运动的最优控制。首先对机器人软启动、软制动进行深入的理论分析，创新性地提出基于参数最优和强化技术控制算法，用若干个分段常值函数构造参数化控制器，将最优控制问题转化为参数优化问题。其次提出外力干扰下的仿人机器人倒地运动分阶段控制方法，设计机器人倒地柔顺控制自适应算法，使其在触地时刻的地面冲击、触地位置和倒地之后的稳定性三个方面达到最优。同时采用约束变换技术对机器人在快速步行与倒地运动之间状态切换进行衔接，让状态函数原点处的约束做平滑处理。采用初始状态筛选集和改进算子自动修正可行方向，加速求解问题寻优过程，最后通过仿人机器人快速步行和倒地连续运动实验验证所提方法的可靠性和有效性。

仿人机器人的运动研究一直是国内外机器人领域里研究的基础，也是难点之一。本项目主要研究仿人机器人运动规划中各种复杂动作规划与设计，同时处理复杂环境的导航和图像识别技术。提出了基于参数最优的控制算法、基于情感的合作路径规划系统、基于FFTW的新的边缘检测图像处理方法和SIFT的3D环境重建算法。所应用的相关技术对仿人机器人进入家庭、服务于人类有着较好的借鉴作用。