Conditional Random Fields for Behavior Recognition of Autonomous Underwater Vehicles

Abstract

This paper focuses on multi-robot teams working cooperatively in an underwater application. Multi-robot teams working cooperatively to perform multiple tasks simultaneously have the potential to be more robust to failure and efficient when compared to single robot solutions. One key to more effective interaction is the ability to identify the behavior of other agents. However, the underwater environment presents specific challenges to teammate behavior identification. Current decentralized collaboration approaches, such as auction-based methods, degrade in poor communication environments. Sensor information regarding teammates can be leveraged to perform behavior recognition and task-assignment in the absence of communication. This work illustrates the use of Conditional Random Fields (CRFs) to perform behavior recognition as an approach to task monitoring in the absence of robust communication in a challenging underwater environment. In order to demonstrate the feasibility of performing behavior recognition of an AUV in the underwater domain, we use trajectory based techniques for model generation and behavior discrimination in experiments using simulated trajectories and real sonar data. Results are presented with comparison of a CRF method to one using Hidden Markov Models.