论文部分内容阅读
在机器人抓取作业时,目标物体的位姿经常发生变化。为了使机器人在运动过程中能够适应物体的位姿变化,提出了一种基于高斯过程的机器人自适应抓取策略。该方法建立了从观测空间到关节空间的映射,使机器人从样本中学习,省去了机器人视觉系统的标定和逆运动学求解。首先,拖动机器人抓取物体,记录物体的观测变量和机器人的关节角度;然后,利用记录的样本训练高斯过程模型,实现观测变量和关节角度的关联;最后,当得到新的观测变量时,通过训练的高斯过程模型得到机器人的关节角度。经过训练后,UR3机器人成功抓取了物体。
The pose of the target object often changes as the robot grabs the job. In order to make the robot adapt to the pose change of the object during the movement, a self-adaptive robot grasping strategy based on Gaussian process is proposed. The method establishes the mapping from observation space to joint space, and makes the robot learn from the sample, eliminating the calibration and inverse kinematics of the robot vision system. First, the robot is dragged to capture the object, and the observed variables of the object and the joint angle of the robot are recorded. Then, Gaussian process model is trained by using the recorded samples to realize the correlation between the observed variable and the joint angle. Finally, when a new observed variable is obtained, Through the training of Gaussian process model to get the robot’s joint angle. After training, the UR3 robot successfully grabbed the object.