Enhancing Trust in Collaborative Assembly through Resilient Adversarial Reinforcement Learning

Collaborative robots (cobots) are designed to improve productivity and safety in industrial settings. However, to be effective Human-Robot Collaboration (HRC) relies heavily on the human operator’s trust in the robotic partner. This study posits that trust is significantly enhanced by the robot's ability to adapt to human behavior, particularly when the human teammate has a behavior unpredictable and outside the box. To achieve this adaptability, we propose an Adversarial Reinforcement Learning (ARL) framework to the activity planning of the robot. The assembly process is modeled as a Markov Decision Process (MDP) on a Directed Acyclic Graph (DAG). The robot learns an assembly policy using an on-policy algorithm, while a simulated human agent acts as an adversary trained with the same algorithm to introduce disturbances and delays. The proposed approach was applied to a simple industrial case study and evaluated on complex assembly sequences generated synthetically. While the ARL-trained robot did not outperform conventional assembly optimization algorithms in terms of task completion time, it guaranteed robustness against human variability, ensuring task completion within a bounded timeframe regardless of human actions. By demonstrating consistent performance and adaptability (Ability) in the face of uncertainty, the robot exhibits characteristics that align with the Ability and Benevolence components of the ABI model of trust, thereby fostering a more resilient and trustworthy collaborative environment.