Model-based controllers are at the heart of many robotic systems. For fixed-wing aircraft in particular, controllers are often driven by models of varying fidelity. Model fidelity selection is an application-specific design choice. On one end of the spectrum, a simple model will be computationally efficient, but at the risk of under modeling. On the other end of the spectrum, a complex model has the potential to better capture system dynamics, but at the risk of increased complexity and potential over modeling. A viable middle ground in model selection is choosing a simple model parametrized by a set of state-dependent constants. Constants can represent unknown or random model parameter to be identified, such as intrinsic system properties or noise terms. The approach taken here is to capture fixed-wing aircraft model uncertainty including noise terms due to wind effects with a set of parameters that can easily be estimated from training data. A technique based on weighted least squares, know as locally weighted models (LWMs), has shown promise in such applications. LWMs strike a balance between simplicity and complexity by representing dynamic systems with a family of locally linear models. To demonstrate the potential of this modeling technique for complex systems, an augmented aircraft model is presented as an example, with a procedure for identifying unknown parameters. The LWM’s predictive capability is compared against a standard non-parametric model as well as flight data collected on a small fixed-wing unmanned aerial vehicle.

This content is only available via PDF.
You do not currently have access to this content.