Design And Implementation Of Robot Arm Automatic Unloading System Based On Machine Vision

With the development of computer technology and robotics,automated operations are becoming more and more common in factories.Although simple mechanical automation equipment and industrial robots have been able to replace manuals for some assembly line work,the flexibility and adaptability of robots are still insufficient,especially in complex scenarios where it is difficult for robots to replace human operations,and automatic unloading is one of the most challenging tasks in logistics and warehousing.The implementation of automatic unloading tasks can reduce the workload of manual operations,improve efficiency,and avoid the occurrence of injuries in unloading.The difficulties in the task of automatic unloading are:first,the "weak structure" of the containers and trailers in the placement of goods,so that the target identification,grasping and handling are difficult;second,improper unloading order may lead to inefficiency,and even the collapse of goods;in addition,in the space of high complexity,the mechanical arm movement planning problem is also One of the difficulties in realizing automatic unloading.Considering the needs and difficulties of automatic unloading tasks,a series of related researches are conducted in this thesis.First,a robotic arm motion planning algorithm based on a safe moving corridor is proposed.The algorithm uses path planning and spatial expansion algorithms to establish a safe movement corridor,and achieves a collision-free smooth trajectory using constraints on the safe movement corridor and a convex optimization algorithm.Second,a robotic arm motion planning algorithm based on reinforcement learning and the TEACHER-STUDENT model is also proposed.The algorithm introduces privileged information and contrast learning loss during training,which not only achieves faster training speed but also gets better motion planning results.Finally,a dataset generation method is designed for the training of the stacked cargo detector.The unloading sequence generation algorithm uses a sequence generation algorithm based on heuristic information to finally obtain the position of the goods that the robotic arm needs to handle each time.Based on the above algorithm,an automatic machine vision-based unloading system is implemented.The system includes three modules:a YOLOv5-based cargo segmentation processing module,a heuristic algorithm-based unloading sequence generation module and a motion planning module.The system contains a more robust target detection algorithm,a more efficient unloading algorithm and a safer motion planning algorithm.The system is a complete automatic unloading system that can be deployed and used end-to-end.To validate the feasibility of the system and test its efficiency,the thesis was deployed and tested on CoppeliaSim virtual environment and real robot platform respectively.The real robot platform uses a UR5 six-degree-of-freedom robotic arm with ROS for control.The experiments show that the system has a stable and efficient automatic unloading capability,and the unloading success rate of the whole system reaches 98.76%,with an unloading speed of 196.7 pieces per hour.