Combined Bi-RRT and Q-Learning path-planning in collaborative environments

In recent years, a significant transformation towards intelligent manufacturing systems has been observed in industry. One of the leading research topics in this field is collaborative robotics, which promotes a synergic interaction between humans and robots. Advantages in ergonomics and production are foreseen with the adoption of collaborative robotics. Avoiding unintended collisions, which would ensure seamless collaboration, is one of the main challenges in improving safety and productivity. This paper focuses on a decision-making strategy that allows the robot to autonomously identify the optimal path to minimize the travel distance between the current configuration and the target while maintaining a safe distance from the human collaborator. The proposed strategy involves the offline generation of a dataset of possible paths within the robot workspace and a Reinforcement Learning-based control strategy, enabling the optimal choice of the subsequent robot configuration. After training and testing in a simulated environment, the optimal policy was validated with an ABB GoFa™ robotic arm, testing different human configurations and paths.