Mobile Grabbing Robot with Target Detection Using Quadric Fitting

Abstract

The recognition and localization of targets, navigation, and mechanical control are key problems for indoor grabbing task. This paper designs a vision-based mobile grabbing robot with a wheeled chassis and a six-axis mechanical arm. The vision perception employs multiple sensors to accomplish different function modules. A two-dimension (2D) map is draw by a lidar and tag codes fixed on the roof is observed by a binocular stereovision to realize a vision-based SLAM. In robot control, the path planning algorithm and kinematics of the mechanical arm determine robot trajectory and grabbing action, respectively. Three-dimension (3D) points colored by an inbuilt camera are acquired by a Kinect v2. For grabbing shape-specified targets, we propose a novel target detection approach, which combines 2D image-based recognition with 3D refined segmentation by quadric fitting. Finally, the feasibility of the robot scheme and the high accuracy of the proposed target detection approach are verified by experiments. The results show that the error of positioning robot is 14.1 cm at the speed of 0.2 m/s and the error of positioning target is 0.53 cm.