Abstract: To adapt to different poses of target object as well as move smoothly,an adaptive strategy that built two Gaussian processes for pose adaption and trajectory adjustment respectively was proposed.The Gaussian process for pose adaption correlated the joint angles of robot with observation variables with linear covariance function,which eliminated the vision calibration and inverse kinematic solutions;the Gaussian process for trajectory adjustment correlated different trajectory points with Gaussian kernel function,which made the adjusted trajectory move smoothly.Grasping experiment of UR3 robot with obstacles proved that the proposed method could make robot adapt to the different poses of target object and obtain smooth joint trajectories.

摘要： 为适应目标物体位姿的变化,实现运动的平滑性,提出一种机器人自适应策略,在位姿适应和轨迹调整两个环节上各建立一个高斯过程模型。位姿适应的高斯过程通过线性协方差函数将观测变量和机器人关节变量关联,避免了视觉系统的校正和机器人运动学逆解;轨迹调整的高斯过程利用高斯核函数计算关节轨迹点之间的协方差,使调整后的机器人轨迹更加平滑。通过UR3机器人在有障碍物下的自适应抓取实验,证明了所提方法既能够适应目标物体的位姿变化,又能得到平滑的关节运动。

关键词: 高斯过程, 自适应策略, 从演示中学习, 高斯核函数, UR3机器人