AN AUTONOMOUS EYE-IN-HAND ROBOTIC SYSTEM FOR PICKING OBJECTS IN A SUPERMARKET ENVIRONMENT WITH NON-HOLONOMIC CONSTRAINT

Designing a robot that can autonomously find and pick objects in a supermarket is the core requirement of an unmanned supermarket. The difficulties of this task include building a robot with both high payload and big workspace and designing a robot system that incorporates the navigation and grasping system. In this paper, we present our robot which includes a high payload chassis, an UR5 manipulator and a suction system. To deal with the non-holonomic constraint and to incorporate the navigation and grasping system, we designed a pipeline called navigation, observation and grasping (NOG). Furthermore, we propose a new grasping pose generation algorithm which utilizes a heuristic cost function to get rid of the effect of the noise points in the point cloud. Jingdong held a competition named Jingdong robotics challenge (JRC) to simulate an unmanned supermarket environment. This competition required the robot to autonomously find and pick 10 different objects among 20 categories of objects and place them into a collection box. We tested our robot in this competition, and the designed robot system grasped 8/10 objects in the first round, 7/10 object in the second round and finally achieved third and first place in two rounds.