Robust Path Planning in GPS-Denied Environments Using the Gaussian Augmented Markov Decision Process

P. Lommel,M. McConley

Abstract:As the field of autonomous robotics continues to mature, the a bility for robots to operate reliably in the presence of noise and incomplete information is becoming increasing ly critical. State estimation techniques such as the Kalman filter explicitly model sensor noise and incomplete i nformation by computing a probability distribution over the possible states. The uncertainty of the posterior p obability distribution is a measure of the expected error of the state estimate, but the challenge is how to incorporat e knowledge of uncertainty into the decision making process. Ignoring the uncertainty that accompanies probab ilistic state estimates during planning can lead to plan failures where the estimate is highly uncertain and likely t o be wrong. In this paper we describe a path planning algorithm that choo ses actions based on both the mean and covariance of the state estimate, efficiently planning in the space of Ka lman filter beliefs. We demonstrate that the planner is robust to potential sensor noise and uncertainty in the ro bot’s pose, achieving significantly better performance in GPS-denied environments compared to shortest-path plan ners.