Infrared Camera Assisted UAV Autonomous Control via Deep Reinforcement Learning
Tao Feng, Manuel Cortez, Yongcan Cao · AIAA Scitech 2021 Forum · 2021
View Video Presentation: https://doi.org/10.2514/6.2021-1121.vid In this paper, we study the problem of implementing a deep reinforcement learning method that controls a physical UAV to navigate from a random initial static state to a desired destination. Different from most existing reinforcement learning applications on UAV control, which heavily depend on software simulators, we directly conduct the training process on a physical UAV. To address the challenges in training with physical UAVs, \textit{i.e.}, resource limitation of on-board chips and the data hungry issue of reinforcement learning, we propose a deep reinforcement learning control framework, and propose a reward engineering approach in the context of the considered task. We also present some basic procedures to build an indoor real drone testing environment with an OptiTrack system. Finally, the testing results of the deep reinforcement learning are presented to demonstrate the performance of the proposed framework.