Markov Decision Processes in Artificial Intelligence

Chapter 13

Autonomous Helicopter Searching for a Landing Area in an Uncertain Environment 1

This chapter presents the application of MDPs to a problem of strategy optimization for an autonomous search-and-rescue helicopter exploring a partially known and uncertain environment in search for a landing zone. Online MDP optimization algorithms were developed and implemented in a real-time aerial robotic architecture, thus providing on-board autonomous bounded-time decision-making capabilities for uninhabited rotorcraft systems. Both theoretical aspects and practical implementation constraints are discussed. This work was part of the ONERA1 RESSAC project [FAB 07]. Figure 13.1 shows the autonomous helicopter exploring an initially unknown environment cluttered with artificial cardboard obstacles. The following functions are performed autonomously on-board the uninhabited rotorcraft: terrain exploration and mapping (perception), online optimization of a conditional exploration strategy (decision), acting according to this strategy, characterization and choice of landing zones (and autonomous landing and take-off). The human operator only intervenes for final validation before landing, or security management during the experimental flights.

13.1. Introduction

The operation of uninhabited systems in partially known dynamic environments requires anytime decision and reaction processes in order to enable the system to deal timely and properly with encountered situations. A number of attempts ...