Intelligent Flight Control

Intelligent Flight Control (IFC) is defined as autonomous, adaptive control algorithms that can find non-trivial solutions to control problems using trivial strategies. There is a worldwide increase in unmanned aviation, but integration of unmanned aircraft in the current airspace is difficult due to safety regulations. While conventional commercial aviation autopilots are not adaptive (the human pilot has to act if something unexpected happens), autonomous systems should be able to to adapt to changing circumstances themselves. This is where Intelligent Flight Control systems fit in.
As an additional benefit, IFC does not need an accurate model of the aircraft to tune the control parameters. This is especially useful when the aircraft is difficult to model, such as the Delfly flapping wing MAV.
Within IFC we focus on Reinforcement Learning (RL) and Approximate Dynamic Programming (ADP) methods. These methods are based on human learning and the concept of building a value function based on rewards obtained from the environment. The value function has a recursive property that is described by the Bellman equation. A policy (state to action mapping) is derived from the value function such that the agent will perform actions that lead it to states with high value. With simple definitions of reward function, complex behavior can be learned.

Applications of IFC

We apply IFC to many different platforms:

• Discrete gridworlds (mainly for testing new algorithms)
• Simulation models
  • Quadrotors
  • Fighter jet
  • Flapping wing UAV’s
• Flying hardware
  • Quadrotors
  •  Flapping wing UAV’s