Learning collaborative team behavior from observation

This paper describes an approach to creating a simulated team of agents through observation of another team performing a collaborative task. Simulated human teamwork can be used for a number of purposes, such as automated teammates for training purposes and realistic opponents in games as well as in military training simulation. Current simulated teamwork representations require that the team member behaviors be manually programmed into the agents, often requiring much time and effort. None of the currently documented techniques for multi-agent learning employ observational learning and a context-aware framework to automatically build agents that replicate the collaborative behaviors observed. Machine learning techniques for learning from observation and learning by demonstration have proven successful at observing the behavior of humans or other software agents and creating a behavior function for a single agent. This technique described here known as COLTS combines current research in teamwork simulation and learning from observation to effectively train a multi-agent system capable of displaying effective team behavior in limited domains. The paper describes the background and the related work by others as well as a detailed description of the learning method. A prototype built to evaluate the developed approach as well as the extensive experimentation conducted is also described. The results indicate success in the selected experiments.