Apprenticeship bootstrapping: Inverse Reinforcement learning in a multi-skill UAV-UGV coordination task

Apprenticeship learning enables learning from human demonstrations performed on tasks. However, acquiring demonstrations in complex tasks where a human expert is not available can be a challenge. In this paper, we propose a new learning algorithm, called Apprenticeship bootstrapping via Inverse Reinforcement Learning using Deep Q-learning (ABS via IRL-DQN), to learn a complex task through using demonstrations performed on primitive sub-tasks. The algorithm is evaluated on an aerial and ground coordination scenario, where an Unmanned Aerial Vehicle (UAV) is required to maintain three Unmanned Ground Vehicles (UGVs) within a field of view of the UAV's camera (FoV). The results show that performance of our proposed algorithm is comparable to that of a human, and competitive to the original IRL using expert demonstrations performed on the composite task. © 2018 International Foundation for Autonomous Agents and Multiagent Systems (www.ifaamas.org). All rights reserved.