Applications of Probabilistic Graphical Models to Diagnosis and Control of Autonomous Vehicles

摘要

We present the main elements of a distributed architecture supporting diagnosis and control of autonomous robots. The purpose of the architecture is to assist the operator or piloting system in managing fault detection, risk assessment, and recovery plans under uncertainty. The architecture is generic, open, and modular consisting of a set of interacting modules including a decision module (DM) and a set of intelligent modules (IMs). The DM communicates with the IMs to request and obtain diagnosis and recovery action proposals based on data obtained from the robot piloting module. The architecture supports the use of multiple artificial intelligence techniques collaborating on the task of handling uncertainty. In this paper we focus on the application of Bayesian modeling to three problems of diagnosis and control of autonomous robots or vehicles. The paper describes and discusses how we use Limited Memory Influence Diagrams (LIMIDs) to represent and solve complex problems of diagnosis and control of ground and underwater robotic vehicles. In particular, we describe how battery monitoring and control problems related to an underwater and a ground vehicle are solved and how a sonar image quality assessment problem related to an underwater vehicle is solved. 1