Towards Lifelong Social Robot Navigation in Dynamic Environments

We describe a work-in-progress approach to solving the problem of robot navigation in dynamically changing, social environments. Our approach employs reinforcement learning informed by a continually updated model that predicts the evolution of the environment, and handles two common scenarios: (1) a person moving within the environment, and (2) static obstacles with positions that change over time. We assess the effectiveness of the approach in a simulated assistive-care application in which a mobile robot supports a person with mild cognitive or physical impairments with simple everyday tasks.