Data Driven Flight Control for Multicopters - A Survey

In the constantly evolving world of multicopters, innovation is the key to unlocking their full potential. An interesting approach involves combining traditional flight control methods like PID, Sliding-Mode-, Cascade- or Model-Predictive-Control with data-driven techniques from current research, forming a concept known as “Data-Driven Flight Control” (DDFC). DDC in general is devided into the darkgray box direct DDC (dDDC) approach and the black box indirect DDC (iDDC) approach. One popular example for dDDFC, is the replacement of linear state-space models with Hankel matrices according to Willems fundamental lemma [1], where entries of the matrices are to be identified through persistently exciting (PE) signals. By synergizing data analysis and advanced control algorithms, dDDC offers a potential way to enhance the precision, adapt-ability, and performance of multicopters and complex systems in aviation. Despite the usefulness of dDDC for UAV applications, not much work has been carried out to develop efficient on-board algorithms. To understand an unknown system, the input and output data to create a Hankel matrix needs to be collected. Normally, this process takes a longer period than necessary to get a bigger matrix. However, solving the optimization problem with this big matrix requires a lot of variables. This makes it difficult to use dDDC in real-time applications or on devices with limited memory as this is the case with UAVs. As the efficiency of on-board computing units increases (we will give an example of a low budget AI-test carrier in the appendix) the application of this methodology is not far from being of practical use in the industry.