Towards Model-Based Robot Behaviour Adaptation: Successful Human-Robot Collaboration in Tense and Stressful Situations

In a demanding environment such as search and rescue, where human-robot collaboration provides potential benefits, understanding complex human behaviour will promote greater success. We propose that robots need to be equipped with embodied models of both human behaviour and the environment. Here we emulate some conditions of a disaster area to test collaboration between a robot and human, with added environmental distraction (noise) and cognitive load. With this setup, we generate empirical data as the basis of human performance modelling, using machine learning techniques. The experiment was repeated in three phases, first to capture baseline data for the basis of the model and later comparison with the second phase, where the robot adapted its working pace based on participants' performance, and the third phase, where the robot instead provided feedback to the participant based on their performance. Results suggest that, in a situation which is tense and stressful for human, having a robot adapting its behaviour towards that of a person can improve the overall success rate of the task at hand.