Reward and Diversity in Multirobot Foraging

摘要

This research seeks to quantify the impact of the
\nchoice of reward function on behavioral diversity in
\nlearning robot teams. The methodology developed
\nfor this work has been applied to multirobot foraging, soccer and cooperative movement. This paper
\nfocuses specifically on results in multirobot foraging. In these experiments three types of reward are
\nused with Q-learning to train a multirobot team to
\nforage: a local performance-based reward, a global
\nperformance-based reward, and a heuristic strategy
\nreferred to as shaped reinforcement. Local strategies provide each agent a specific reward according
\nto its own behavior, while global rewards provide
\nall the agents on the team the same reward simultaneously. Shaped reinforcement provides a heuristic reward for an agent's action given its situation.
\nThe experiments indicate that local performance-based rewards and shaped reinforcement generate
\nstatistically similar results: they both provide the
\nbest performance and the least diversity. Finally,
\nlearned policies are demonstrated on a team of Nomadic Technologies' Nomad-150\t robots.