Online PID Self-Tuning using an Evolutionary Swarm Algorithm with Experimental Quadrotor Flight Results

Abstract

PID is one of the most common control laws in existence, used extensively across almost every engineering discipline. One potential application of PID is in unmanned air vehicle (UAV) flight. A common problem faced by control law designers is the tuning of such controllers which can be a very time consuming and sometimes difficult process especially on hardware. Various past techniques have considered how PID can be self-tuned. For example, genetic algorithms have been shown to perform offline optimization, which is normally a very long and unviable procedure. In this research we consider the use of online optimization to allow self-tuning of PID. This is done using an ABC colony technique which, compared to genetic algorithms, offers greater simplicity and excellent optimization performance. Applying the online technique has shown to be a difficult task due to the lack of reliability of the online fitness function; the complexity is even more exacerbated in a real-life UAV with a complex PID control architecture. Some important enhancements to the online ABC tuning (in particular the fitness function) have been developed in order to obtain better PID constants. Results on a motion-capture testbed demonstrate that the algorithm combined with the improved fitness function has the ability to select optimal P, I and D values on a quadrotor in real-time; this has never been achieved before experimentally.