Exploration and Monitoring With Gaussian Processes for Path Planning

Please see the linked paper for full details! This paper was completed as a two week custom project for EECS 106B, Robotic Manipulation and Interaction, at UC Berkeley.

Abstract — Robotic exploration and monitoring systems must operate in diverse, unknown, and dynamic environments, posing challenges for pre-planned routes and distributing temporal resources. Researchers have developed statistical methods to estimate functions using noisy measurements. In particular, Gaussian Processes (GPs), yield a model that also estimates the uncertainty in its estimate. In this paper, we propose an approach to exploration and monitoring in which the robotic system optimizes its route to decrease GP posterior uncertainty via Value Iteration over a discretized occupancy grid. We hypothesize that this goal, over reducing route cycle time, will improve map construction rate and decrease steady state uncertainty of dynamic markers.