Task and Motion Planning of Service Robot Arm in Unknown Environment Based on Virtual Voxel-Semantic Space

Abstract

A task and motion planning method for service robot arm based on 3-D voxel-semantic maps is proposed, which can realize virtual environment mapping, manipulator planning, and grasping tasks in unknown environments. First of all, a complete point cloud scene is obtained and spliced. Mask region-based convolutional neural network (RCNN) network is used to complete object detection and instance segmentation. A voxel-semantic hybrid map composed of 3-D point cloud, semantic information, and 3-D computer aided design (CAD) model is constructed. Second, an improved A* algorithm is proposed to plan the optimal path of robot arm end-effector. The Bezier curve interpolation is introduced to obtain the smooth trajectory. Third, the grasping poses of the robot gripper corresponding to different geometries are explored. Semantic-driven spatial task planning is achieved by decomposing robotic arm pick and place tasks. Finally, the effectiveness and rapidity of the proposed algorithm are verified in virtual space and real physical space, respectively.