Deep Reinforcement Learning (DRL) for UAV Control in Gazebo Simulation Environment
2018-09-12
1 System Introduction
Deep Q-Network (DQN) is utilized for UAV altitude control(hovering) and Gazebo is used as the dynamic simulator to interact with the drone.
Github: PX4-Gazebo-Simulation
2 Our Intention
Proportional-integral-derivative (PID) controllers are used as low-lever controllers to achieve desired thrust and three desired Eular angles, and DQN is used for position control.
PX4 is an open-source autopilot and is widly used for drones.
3 Other requirements: ROS and MAVROS
4 Project Overview
5 Constraints and Reward
