Unified Safety-Aware Physical Human-Robot Collaborative Controller with Online Continuous Learning and Personalized Tuning

The growing demand for tailor-made products is pushing industries to develop new strategies that increase flexibility and efficiency in manufacturing. Rather than replacing humans entirely, experts emphasize the essential role of collaborative robots in enabling this transition. As a result, improving control strategies for physical Human-Robot Collaboration (pHRC) has become increasingly important. However, state-of-the-art approaches commonly address safety, intention detection, and adaptability to users and tasks separately. This work presents an enhanced control strategy for pHRC that simultaneously improves safety, human intention estimation, and robot adaptation in real time. To this end, the method leverages: i) fuzzy logic with embedded safety rules; ii) Artificial Neural Networks (ANNs) to model the pHRC dynamics; iii) a Q-Learning framework to compute optimal control parameters (i.e., setpoint and damping); and iv) user-centric control parameter tuning. The approach is experimentally validated using a Franka EMIKA Panda robot in two scenarios: (i) comparison with a state-of-the-art method, and (ii) assessment of personalized tuning, evaluating how user-driven adjustments influence the robot's responsiveness and performance. Achieved results show the effectiveness of the proposed controller.