Where is My Favourite Toy? Inferring the Mental States of Users in False Belief Understanding

摘要

The increasing prevalence of social robots in today's world has made it crucial to have autonomous systems that can interact with humans and adjust to their behaviour. Doing so requires a deep understanding of the human mind, including complex mental states such as beliefs and preferences. To tackle this issue, we have developed a model that can identify false beliefs using the principles of Theory of Mind (ToM), a unique human cognitive mechanism that attributes mental states to others. False belief understanding has always been the primary benchmark used to evaluate ToM in psychology, and this still remain true when testing it in machine systems such as robots. Our model is a modified version of the Bayesian Theory of Mind (BToM), a probabilistic model that reasons on agents' mental states regarding their interactions within the environment. To test the model's performance, we set up a complex assistive scenario with a robot and two human agents playing with toys. In this scenario, the model serves as the cognitive component of the robot, responsible for organising a room with toys while considering the preferences and beliefs of the agents regarding the toys' locations. We have provided results to demonstrate the model's performance in different conditions. Additionally, we have used the Unity Engine as a platform to simulate the cleaning scenario and show the robot's role in such situations.