Using Imitation to Build Collaborative Agents

Abstract

The article presents an approach to learn collaborative strategies among multiple agents via imitation. Imitation-based learning involves learning from an expert by observing the demonstration of a task and then replicating it. This mechanism makes it convenient for a knowledge engineer to transfer knowledge to a software agent. This article applies imitation to learn not only the strategy of an individual agent, but also the collaborative strategy of a team of agents to achieve a common goal. The article presents an imitation-based solution that learns a weighted naïve Bayes structure, whereas the weights of the model are optimized using Artificial Immune Systems. The learned model is then used by agents to act autonomously. The applicability of the presented approach is assessed in the RoboCup Soccer 3D Simulation environment, which is a promising platform to address many complex real-world problems. The performance of the trained agents is benchmarked against other RoboCup Soccer 3D Simulation teams. In addition to performance characteristics, the research also analyzes the behavioral traits of the imitating team to assess how closely they are imitating the demonstrating team.