A Method of Tight Placement for Robotic Dense Packing.

Recent advances in the field of robotics, sensor technology and artificial intelligence, cause the perception and policy of robots to be more and more efficient, which motivates the process of packing automation in warehouses. Nevertheless, compared with manual operation, current robot packing methods tend to use a minimalistic suction cup as the end effector, which lacks flexibility and always needs subsequent push and pull actions to ensure compactness, resulting in low efficiency. Moreover, current methods mostly lack the ability of external force/torque perception, bringing much uncertainty and danger in such a contact-rich task. In this paper, we design a framework based on force/torque perception and compliance control to improve the performance in robustness and safety. Meanwhile, with the proposed compliance controller, a method of tight placement for dense packing is proposed to place cubes in the bin successfully and tightly, without additional actions as assistance. Specifically, we focus on the situation where the robotic arm has been visually guided near the desired placement position by predefined placement order and the pose of the bin, where the effect of errors caused by vision are considered in our algorithm. Finally, the proposed pipeline is evaluated in abundant contrast experiments and demonstrated to be relatively robust and accurate under errors and uncertainties, compared with prior methods.