LAMP: Learning a Motion Policy to Repeatedly Navigate in an Uncertain Environment

Mobile robots are often tasked with repeatedly navigating through an\nenvironment whose traversability changes over time. These changes may exhibit\nsome hidden structure, which can be learned. Many studies consider reactive\nalgorithms for online planning, however, these algorithms do not take advantage\nof the past executions of the navigation task for future tasks. In this paper,\nwe formalize the problem of minimizing the total expected cost to perform\nmultiple start-to-goal navigation tasks on a roadmap by introducing the Learned\nReactive Planning Problem. We propose a method that captures information from\npast executions to learn a motion policy to handle obstacles that the robot has\nseen before. We propose the LAMP framework, which integrates the generated\nmotion policy with an existing navigation stack. Finally, an extensive set of\nexperiments in simulated and real-world environments show that the proposed\nmethod outperforms the state-of-the-art algorithms by 10% to 40% in terms of\nexpected time to travel from start to goal. We also evaluate the robustness of\nthe proposed method in the presence of localization and mapping errors on a\nreal robot.\n