Agile Quadcopter Flight
Deep reinforcement learning for quadcopter racing.
Deep reinforcement learning (deep RL) has shown tremendous success in its application to robotic tasks. In this research, I set out to investigate its application in agile quadcopter flight in the setting of drone racing. Additionally, I seek to explore training a deep RL online path planner that enable the quadcopter to fly through unseen race tracks.
Agile quadcopter flight in the square racetrack.
Agile quadcopter flight in the S racetrack.
Inspired by the curriculum training in Rudin et al., I aimed to develop a deep RL model for online quadcopter path planning by training it to incrementally fly through more and more gates. As a first step, I trained the quadcopter to fly through two consecutive gates in different placements (shown in the video on the right).