A Behavioural Conditioning Approach to Autonomous Flight Control Design

Unmanned Air Vehicles are aircraft designed for autonomous operation in realistic environments. The condition of autonomy restricts major mission planning and flight control to on-board processing. This paper presents a novel control architecture, CEDAR, that is structured from reactive controllers to satisfy real-time operation. Adaptive circuitry provides for compensation of design errors and allow the control system to learn skills through experience. Reinforcement learning is the basis for behavioral conditioning and planning. Simulation results are presented which demonstrate the on-line training capabilities of a sailplane autopilot. An overview of the architecture covers the basic operation and related works.