Agile Flight with Deep Reinforcement Learning
Miss Xie Yuhan
MPhil candidate in the Mechanical Engineering Dept.
Date & Time
Tuesday, 26 April 2022
Quadrotors are agile. Unlike most other machines, they can navigate through complex structures, fly through damaged buildings, and reach remote locations inaccessible to other robots. To date, only expert human pilots have been able to fully exploit their capabilities. Autonomous operation with onboard sensing and computation has been limited to low speeds. Current state-of-the-art works addressed agile quadrotor navigation by splitting the task into a series of consecutive subtasks: perception, map building, planning, and control. Although this approach has proven successful at low speeds, the separation it builds upon can be problematic for high-speed navigation in cluttered environments. The subtasks are executed sequentially, leading to increased processing latency and a compounding of errors through the pipeline. Recent researches exploit deep learning techniques to unlock autonomous agile flight in complex environments, which will be introduced in this seminar.
Robotics and Control