Learning executable agent behaviors from observation

We present a method for learning a human understandable, executable model of an agent's behavior using observations of its interaction with the environment. By executable we mean that the model is suitable for direct execution by an agent. Traditional models of behavior used for recognition tasks (e.g., Hidden Markov Models) are insufficent because they cannot respond to input from the environment. We train an Input/Output Hidden Markov Model where the output distributions are mixtures of learned low level actions and the transition distributions are conditional on features calculated from the agent's sensors. We show that we are able to recover both the behavior and humanunderstandable structure of a simulated model inspired by animal behavior studies. We also present a novel training method that combines multiple EM trials through discrete optimization.

• Type: Conference
• Year: 2006
• Where: ATAL
• Authors: Andrew Guillory, Hai Nguyen, Tucker R. Balch, Charles Lee Isbell Jr.
• Keywords: ATAL 2006, Behavior, Hidden Markov Model, Intelligent Agents, Model