Markov Decision Processes in Artificial Intelligence

Chapter 14

Resource Consumption Control for an Autonomous Robot 1

This chapter presents a robotic application. The objective is to control a robot’s resource consumption using a Markov Decision Process. An autonomous rover has to explore a number of spots in order to gather information. The robot’s environment is inaccessible for human beings. In this context, embedded consumable resources like memory, power or time are bounded. At a certain time, a mission is sent to the rover, which can gather different pieces of information in the environment. The rover must do its best to gather as much information as it can before having consumed the resources. All tasks must be performed in several steps, but there are various ways to perform them. Generally, there are not enough resources to complete the whole mission. This chapter describes a high-level controller that selects the best way to consume resources while performing the tasks during the mission as well as possible. This selection takes the importance of each task into account but also the resources required by the rover. The resource consumption control is the motivating point of the chapter. [BRE 02, NUL 05] also propose approaches to plan a mission while taking consumable resources into account, but, in contrast with their work, we do not present a planner or a task organizer, just a resource controller. Additionally, this chapter presents the case of a single consumable resource (the remaining time), but it is possible to ...