基于高斯混合模型的工业机器人适应性抓取

摘要

在机器人抓取物体的过程中,机器人要调整自身的位姿,以适应物体位姿的变化.提出了一种基于高斯混合模型的适应性抓取方法,实现了机器人在较大工作区域中对物体的抓取.该方法采用高斯混合模型进行建模,构建物体的观测变量与机器人关节变量之间的映射关系.机器人抓取物体时,首先通过相机获取物体的观测变量,分别计算各个高斯分布下该观测变量的生成概率,选取后验概率最大的分布对应的高斯过程回归得到适应性的机器人关节角度.实验结果表明,采用高斯混合模型建模,比采用单一的高斯过程建模能够使机器人更好地实现适应性抓取.%In order to grasp an object, the robot need adjust its pose to adapt to the change of the object pose. An adaptive grasping method based on Gaussian mixture model is proposed, which realizes the adap-tive grasping of the robot in the workspace. A Gaussian mixture model is used to construct the mapping be-tween the observation variables of the object and the robot joint variables. When the robot grasp an object, firstly the observation variables of the object are obtained through the camera. Then, the generating probabil-ities of the observation variables under each Gaussian distribution are calculated, and the Gaussian process regression corresponding to the distribution which has the highest posterior probability is selected to obtain the adaptive robot joint angles. The experimental results show that modeling by a Gaussian mixture model can make the robot grasp adaptively better than a single Gaussian process. The robot vision system calibra-tion, kinematics inversion and planner design are eliminated.