Abstract
- Ensuring safety in shared human-robot workspaces requires dynamic solutions to address unpredictable human behavior. Traditional fixed safety zones or reactive methods often fail to adapt in real time, leading to inefficiencies and risks. This paper proposes a framework that integrates deep reinforcement learning and long short-term memory-based trajectory prediction to dynamically manage safety zones. The system uses a laser scanner for intruder detection, deep reinforcement learning agents for adaptive zone management, and a long short-term memory network for motion prediction. By proactively adjusting safety margins and auxiliary points, the framework optimizes coverage while minimizing workspace restrictions. Simulations with a robotic arm and laser scanner validate the approach, showing improved safety, adaptability, and operational efficiency over conventional methods.