Autonomous UAV Trail Navigation with Obstacle Avoidance Using Deep Neural Networks

This paper proposes a vision-based bike trail following approach with obstacle avoidance using CNN (Convolutional Neural Network) for the UAV (Unmanned Aerial Vehicle). The UAV is controlled to follow a given trail while keeping its position near the center of the trail using the CNN. Also, to return to the original path when the UAV goes out of the path or the camera misses the trail due to disturbances such as wind, the control commands from the CNN are stored for a certain duration of time and used for recovering from such disturbances. To avoid obstacles during the trail navigation, the optical flow computed with another CNN is used to determine the safe maneuver. By combining these methods of i) trail following, ii) disturbance recovery, and iii) obstacle avoidance, the UAV deals with various situations encountered when traveling on the trail. The feasibility and performance of the proposed approach are verified through realistic simulations and flight experiments in real-world environments.