With the prevalence of AI and robotics, autonomous systems are very common in all aspects of life. Real-world autonomous systems must deal with noisy and limited sensors, termed partial observability, as well as potentially other agents that are also present (e.g., other robots or autonomous cars), termed multi-agent systems. We work on planning and reinforcement learning methods for dealing with these realistic partial observable and/or multi-agent settings. The resulting method will allow agents to reason about, coordinate and learn to act even in settings with limited sensing and communication.
Lab for Learning and Planning in Robotics