Deep Reinforcement Learning-Based Interaction Methods for Human-Collaborative Robots in Industrial Environments

Deep reinforcement learning (DRL) has proven to be a powerful tool for optimizing decision-making in complex, dynamic scenarios. In this study, a safety-aware DRL framework is proposed to enhance human-robot collaboration (HRC) in industrial environments, addressing key challenges such as real-time adaptability, collision avoidance, and intention prediction. The framework incorporates a dynamically adjusted reward function that balances task efficiency and safety by penalizing risky actions and rewarding efficient behaviors. Additionally, a long short-term memory (LSTM)-based model is integrated to predict human intentions, enabling the robot to proactively adapt to human behavior. Risk assessment is achieved through a quantitative function combining relative distance and velocity, ensuring safe navigation in dynamic environments. The framework was evaluated in simulated industrial scenarios, showcasing its effectiveness in both safety and performance. Results demonstrated a $72 \%$ reduction in collision rate and a $25 \%$ improvement in task completion time compared to traditional rule-based methods, illustrating the robustness of the proposed framework in optimizing robot behavior in unpredictable and high-dimensional environments. These findings confirm the potential of DRL approaches to significantly advance collaborative robotics in industrial automation.