人臂可达运动手部轨迹规划与臂构型学习

Hand Motion Planning and Arm Configuration Learning for Human Arm Reachable Motion

应用最小冲击优化理论与多层感知器建立虚拟数字人可达运动模型,研究了人臂可达运动中臂构型与末端手部轨迹的规划、复现问题,并与最小二乘法的轨迹进行了对比研究。搭建三维红外运动数据采集系统,采集约束下的人臂可达运动日常行为数据并分析人臂运动机理,基于最小冲击优化理论对人臂可达运动末端手部轨迹进行规划;建立人臂运动学模型,采用多层感知器的学习方法学习人臂可达运动中的拟人臂构型;最后建立人臂末端轨迹和臂构型数学模型,并搭建和显示虚拟数字人体三维模型,将人臂可达运动中末端轨迹与臂构型相结合,驱动虚拟数字人臂可达运动生成。发现使用多层感知器学习的人臂可达运动中人臂构型误差平均误差MAE为0.0532,均方根误差RMSE为0.0031;末端手部在约束可达运动中路径近似为直线,且速度大致呈单峰钟形曲线。结果表明:最小冲击优化理论的末端手部轨迹规划与多层感知器学习的人臂构型相结合可以很好地反映人臂可达运动机理,是一种研究约束下人臂可达运动复现的有效方法。

Applying the theory of minimum impact optimization and multilayer perceptions to establish a virtual digital human reachable movement model,this paper conducted a study on the arm configuration and the end hand trajectory planning and reproduction problems in the reachable movement of the human arm,and compared the trajectory of the least square method.Setting up a three-dimensional infrared motion data acquisition system,the paper collected daily behavior data of human arm reachable motion under constraints,analyzed the mechanism of human arm motion,and planed the hand trajectory at the end of human arm reachable motion based on the optimization theory of minimum impact;then,established human arm motion Learning model by using the learning method of multi-layer perceptron to learn the anthropomorphic arm configuration in the reachable motion of the human arm;and finally established the human arm end trajectory and the mathematical model of the arm configuration,and built the virtual digital human three-dimensional model.In the reachable motion,the terminal trajectory is combined with the arm configuration to drive the virtual digital human arm to generate the reachable motion.It is found that the human arm configuration error MAE in the reachable motion of the human arm using the multi-layer perceptron is 0.0532,and the RMSE is 0.0031;the path of the end hand in the restrained reachable motion is approximately a straight line,and the speed is roughly a single-peak bell curve.The results indicate that the combination of the end hand trajectory planning of the minimum impact optimization theory and the human arm configuration learned by the multilayer perceptron can well reflect the human arm reachable motion mechanism,which is an effective method to study the reproduction of human arm’s reachable motion under restraint.