The paper, ‘Reaching the Limit in Autonomous Racing: Optimal Control versus Reinforcement Learning’ by Yunlong Song et al., presents a study comparing the effectiveness of Reinforcement Learning (RL) and Optimal Control (OC) in the context of autonomous drone racing. It demonstrates that a neural network controller trained with RL outperforms those using OC. While OC decomposes the problem into planning and control with a trajectory, RL optimizes a task-level objective and copes with model uncertainty, allowing for more robust control responses. This study pushes agile drone performance to its limit, achieving peak acceleration greater than 12 g and peak velocity of 108km/h. The findings present a milestone in agile robotics.
Publication date: 18 Oct 2023
Project Page: https://youtu.be/HGULBBAo5lA
Paper: https://arxiv.org/pdf/2310.10943
