Drones Chasing Drones: Reinforcement Learning and Deep Search Area Proposal

Abstract

Unmanned aerial vehicles (UAVs) are very popular and increasingly used in different applications. Today, the use of multiple UAVs and UAV swarms are attracting more interest from the research community, leading to the exploration of topics such as UAV cooperation, multi-drone autonomous navigation, etc. In this work, we propose two approaches for UAV pursuit-evasion. The first approach uses deep reinforcement learning to predict the actions to apply to the follower UAV to keep track of the target UAV. The second approach uses a deep object detector and a search area proposal (SAP) to predict the position of the target UAV in the next frame for tracking purposes. The two approaches are promising and lead to a higher tracking accuracy with an intersection over union (IoU) above the selected threshold. We also show that the deep SAP-based approach improves the detection of distant objects that cover small areas in the image. The efficiency of the proposed algorithms is demonstrated in outdoor tracking scenarios using real UAVs.

Keywords

• unmanned aerial vehicles
• tracking
• deep learning
• reinforcement learning
• drone pursuit-evasion