This project examines the application of reinforcement learning in order to automate the tuning of electronic speed controller (ESC) parameters, focusing on utilizing Field-Oriented Control (FOC) optimization. A Proximal Policy Optimization (PPO) model is implemented to adjust a subset of control parameters while operating with a ESC coupled to a physical test bench. The test bench includes a motor coupled to a dynamometer, with efficiency and power factor measurements as primary metrics. The results show that the PPO agent was able to improve the overall efficiency while operating in steady-state. Certain limitations have been observed. These include sensitivity to test case characteristics and challenges arising from the lack of dynamic interaction.
Lyckman et al. (Wed,) studied this question.