Collaborative Pushing and Grasping of Tightly Stacked Objects via Deep Reinforcement Learning

Directly grasping the tightly stacked objects may cause collisions and result in failures,degenerating the functionality of robotic arms.Inspired by the observation that first pushing objects to a state of mutual separation and then grasping them individually can effectively increase the success rate,we devise a novel deep Q-learning framework to achieve collaborative pushing and grasping.Specifically,an efficient non-maximum suppression policy(PolicyNMS)is proposed to dynamically evaluate pushing and grasping actions by enforcing a suppression constraint on unreasonable actions.Moreover,a novel data-driven pushing reward network called PR-Net is designed to effectively assess the degree of separation or aggregation between objects.To benchmark the proposed method,we establish a dataset containing common household items dataset(CHID)in both simulation and real scenarios.Although trained using simulation data only,experiment results validate that our method generalizes well to real scenarios and achieves a 97%grasp success rate at a fast speed for object separation in the real-world environment.

Novel objects 3-D dense packing through robotic pushing

Robotic picking and placing systems can increase the flexibility of packing objects in a box and are attractive in many fields.However,owing to inevitable uncertainties in both the picking and placing stages,the objects cannot be placed at desired positions accurately and hence cannot be packed densely.This paper presents an additional pushing action that maximizes the packing density;i.e.,after being released from the robot end,the object is moved by robotic pushing actions to arrange the packing densely.The robotic pushing strategy is determined through a deep reinforcement learning algorithm.The idea is to compress the objects toward a corner to improve the volume utilization rate by minimizing the result of a heuristic score.The learning process is implemented in simulation and the trained network is transferred to a robot system directly.Simulations and experiments are presented for the packing of regular and irregular objects to verify the proposed method.