Manipulator Collision Avoidance Path Generation Algorithm Considering Predicted Human Path

In a collaborative robot system, humans and robots can share a work area to improve work efficiency. However, since there is a high risk of human injury when a collision occurs within the workspace, robots must be able to avoid collisions with humans while performing tasks. Existing studies generate a path by assuming a constant speed for a moving person, resulting in a straight line or arc that cannot reflect complex human motor characteristics. In this paper, we propose an algorithm that predicts the path of human motion with a network using a Gated Recurrent Unit (GRU) and generates an Artificial Potential Field (APF) that reflects the predicted path while considering the future collision potential. Within the created APF, we generate a collision-avoidance path that reaches the destination quickly.