Objectives
The objective of this project is to investigate various methods for the control of an unmanned morphing aerial vehicle. Research topics may include: optimal control strategies, estimation of the vehicle states, and data fusion of multiple sensor types.
Background
A morphing aircraft has been developed with additional degrees of freedom that are not found in conventional aircraft. These new degrees of freedom will allow the aircraft to perform new maneuvers, such as a perching maneuver. The capability to perform the perching maneuver will enable the aircraft to land in new situations and locations for increased mission performance. A goal of this project is to autonomously land the aircraft through the perching maneuver.
Research
Several investigations have been conducted into various aspects of the control of the morphing aircraft. Information returned by the sensors must be properly combined to allow the aircraft to operate as desired. This information may be used in many ways, for instance in the determination of the states of the aircraft or the local environment. Knowledge of these states is important for control of the aircraft, especially during the landing phase. Simulations of the aircraft are used to evaluate the performance of methods investigated.
Figure 1: General overview of system components
Many types of sensors may be used to return information about the aircraft and environment. Measurements may include information regarding: airspeed, barometric pressure, gravity vector, accelerations, rotations rates, position, and velocity. The sensors used may not all be synchronous, i.e. data may be returned at different rates. Often not all of the system's states can be measured directly and must be estimated. There measurements may contain inaccuracies, and it is important that these errors be overcome to combine the data properly to arrive at an accurate description of the system's states. A common estimator that may be used in such situations is the Kalman Filter (KF), which creates an estimate of the system's states by balancing the relative accuracies of the sensor measurements and assumed system's dynamics. Some investigations into this issue were conducted to determine how various sensors affected the accuracy of the estimates. Data were returned at multiple rates to simulate real-world behavior.
Figure 2: Simulation to compare estimation using multiple types of sensors
The standard KF method operates on matrices, and is therefore a linear estimator. An aircraft is a nonlinear system, and in some cases a linear estimator may not be able to describe the vehicle with sufficient performance. Alternative estimators may be used to reduce estimation errors. Weighted modal methods are often used to more accurately describe system responses, such as in vibration response, and a related technique can also be used in estimation. Multiple models, which each describe different dynamics, can be combined to produce estimates of the system that will help account for higher-order effects not encapsulated by a single system model. This method was investigated for controlling the aircraft. Several estimates of the system were created simultaneously using a collection of models, and these estimates were compared with the sensor measurements to calculate the weights that were then used to combine the individual estimates to form a global estimate of the system. Control strategies can also be applied in a similar manner, and the number and selection of models used affects the performance.
Figure 3: Comparison of errors in forward position estimates using multiple model method. The diagrams on the left show the locations in the maneuver where the models were selected
Various control methods are also available to direct the aircraft to follow a desired maneuver. A common method of feedback control is to use a proportional-integral-derivative (PID) method, in which gains are set to regulate the control inputs in response to deviations between the actual system and the reference (desired) values. Another method would be to use an optimal method to determine the weights that are applied to produce the control signals. The linear quadratic regulator (LQR) method could be used for this purpose. As is the case with estimation, more advanced control methods also require increased computational resources. Linear methods will be easier to implement, however may not provide sufficient accuracy to allow the aircraft to perform a desired maneuver.
Publications
- Hurst, A., Wickenheiser, A. and Garcia, E. "Localization and Perching Maneuver Tracking for a Morphing UAV", Position Localization and Navigation Symposium, May 5-8, Monterey, CA, 2008. (PDF)
- Hurst, A. and Garcia, E. "Towards Automated Landings of a Morphing UAV", 18th International Conference on Adaptive Structures and Technologies, October 3-5, Ottawa, ON, 2007. (PDF)