Energy-constrained Active Exploration with Incremental Perception

An active exploration framework that optimizes target search under energy constraints given no prior information. We consider a robot with an incremental-resolution symbolic perception deployed in an environment with known geometry and unknown semantics. Casting this energy-constrained sequential decision-making problem as a flow maximization problem, we leverage tools from optimization and graph theory to efficiently plan online as the robot’s knowledge about the semantics of the surroundings is updated.