In this paper, we address an automated gain tuning method using Asynchronous Advantage Actor-Critic (A3C) reinforcement learning approach. A quad-rotor Unmanned Aerial Vehicle (UAV) with nonlinear geometric tracking controller is introduced to test our approach. In the geometric controller, two attitude gains must be provided appropriately to achieve stable error dynamics. To ease the difficulties while optimizing the controller performances, such as minimizing tracking error together with reducing control energy, we made Reinforcement Learning (RL) agents to substitute the entire gain tuning process. By training the RL agents with multiple quad-rotor configurations, we were not only able to reduce our efforts putting into the gain tuning by the trial-and-error methods, but also able to deal with the parameter changes by constructing an adaptive structure.

中文翻译：

---

基于A3C的四旋翼几何姿态控制器的自动增益调谐方法

在本文中，我们使用 Asynchronous Advantage Actor-Critic (A3C) 强化学习方法解决了一种自动增益调整方法。引入了具有非线性几何跟踪控制器的四旋翼无人机 (UAV) 来测试我们的方法。在几何控制器中，必须适当提供两个姿态增益以实现稳定的误差动态。为了在优化控制器性能的同时缓解困难，例如最小化跟踪误差以及降低控制能量，我们使用强化学习 (RL) 代理来替代整个增益调整过程。通过使用多个四旋翼配置训练 RL 代理，我们不仅能够减少通过试错法进行增益调整的努力，