基于神经网络的仿人跑步机器人步态规划

A Neural Network Based Running Gait Planning Method for Humanoid Robot

为了快速生成仿人机器人跑步运动轨迹,研究了一种用于仿人机器人跑步步态生成的步态规划器。采用三维弹簧倒立摆模型描述跑步过程中仿人机器人质心运动规律,奔跑时机器人质心轨迹及落脚点位置可以由四个步态参数来确定,从而将步态规划问题转化成步态参数优化问题,求解了500余种不同运动状态下的步态参数。建立了基于三层BP神经网络的步态规划器,将优化结果作为训练样本训练神经网络。用上述规划器实现了仿人机器人跑步步态规划并对规划结果进行了仿真验证。研究结果表明,基于BP神经网络的步态规划器可以实现步态参数的快速计算,生成的跑步步态逼真;提出的跑步运动步态规划方法可行,为仿人机器人实时轨迹生成提供了一种解决方法。

In order to generate a running trajectory for a humanoid robot rapidly,a real-time gait planner for humanoid robot was researched.Firstly,a three-dimensional spring loaded inverted pendulum(3 D-SLIP)model was used to describe the motion of center of mass of a running humanoid robot.The mass center trajectory of humanoid robot and the touchdown point of the support foot can be determined by four separate parameters.So,the gait planning problem was transformed into a parameter optimization problem.More than 5 hundred groups of gait parameters were obtained through optimization.Then,a gait planner was established based on a three-layer BP neural network.Several hundred of training samples from optimization results were used to train the gait planner successfully.Finally,many groups of gait parameters were generated by the planner and simulation was done to verify the planning result.The research results show that the gait planner based on BP neural network can generate the gait parameters accurately and quickly,and the running gaits are very natural and vivid.Therefore,the proposed method for running gait planning is feasible,which provides a solution for real-time trajectory generation for humanoid robot.