Real-Time Collision Avoidance in a Dynamic Environment for an Industrial Robotic Arm

This paper proposed learning from demonstration control policy that permits an industrial robotic arm to generalise in an unstructured and dynamic environment. The approach combines a probabilistic model with a reactive approach that learns the cost function of an unknown state of the environmental constraints. The approach redefines the robot’s behaviour towards generating a trajectory that satisfies the task and scene constraints. In the end, experiments on a real industrial robotic arm were presented to show how the proposed approach works and facilitates enhanced human-robot coexistence.