Environment Model Adaptation for Autonomous Exploration

Abstract

This thesis proposes adapting a mobile robot’s environment model as a means of increasing the speed at which it is able to explore an unknown environment. Exploration is a useful capability for autonomous mobile robots that must operate outside of controlled factories and laboratories. Recent advances in exploration employ techniques that compute control actions by analyzing information-theoretic metrics on the robot’s map. Information-theoretic metrics are generally computationally expensive to evaluate, ultimately limiting the speed at which a robot is able to explore. To reduce the computational cost of exploration, this thesis develops an informationtheoretic strategy for simplifying a robot’s environment representation, in turn allowing information-based reward to be evaluated more efficiently. To remain effective for exploration, this strategy must adapt the environment model in a way that sacrifices a minimal amount of information about expected future sensor measurements. Adapting the robot’s map representation in response to local environment complexity, and propagating the efficiency gains through to planning frequency and velocity gives rise to intelligent behaviors such as speeding up in open expanses. These methods are used to demonstrate information-theoretic exploration through mazes and cluttered indoor environments at speeds of 3 m/s in simulation, and 1.6 m/s on a ground robot.