Research on collision detection algorithm of robot arm based on multi-view object detection

Aiming at the difficult problem of collision detection in the trajectory planning process of the robotic arm in an unstructured environment, a multi-view object detection algorithm based on geometric collision detection and deep learning is proposed. Three global cameras are deployed in the robot working environment to ensure the integrity of the environmental observation. By pre-training Mask R-CNN, the obstacles in the observed image are detected, and the obstacle circular encircle box is constructed. The geometric collision detection model is used to determine the occupying relationship between the robot arm and the obstacle in the image plane. The test results in the simulation environment show that the proposed method not only has high detection accuracy but also fast detection efficiency, and realizes the collision detection of the robot arm in the unstructured environment.