Autonomous Navigation of a Quadrotor in Unknown Environments

In this paper, the perception, planning, and control of a quadrotor are studied to navigate the quadrotor through unknown confined environments. To deal with the limited sensing and computation capability of the quadrotor, the perception, planning, and control are designed in a coupled manner. The basic idea is to design a method that is less demanding on perception and planning by leveraging the Model Predictive Control (MPC). More specifically, the waypoints and trajectories are generated in real-time using the point-cloud information from a range sensor. The trajectories are generated with respect to a moving reference frame, and the constantly regenerated trajectories act as feedback to guide the quadrotor. Then, a model predictive controller is introduced for trajectory tracking. With the coupled design, no map of the environment needs to be built, and the position of the quadrotor is not needed. Multiple simulations are conducted in the ROS and gazebo environment. The results show that the quadrotor can navigate in unknown, confined environment successfully.