An Edge AI based Robot System for Search and Rescue Applications

In this work, we propose an edge AI based robot system that contains drones and multi-legged robots for search and rescue applications. To accurately search for survivors in real-time, we integrate Tiny-YOLO into the drone design. Instead of adopting a microprocessor usually used in a robot, the FPGA device is adopted as the main hardware computing architecture of the multi-legged robot. A resource-efficient quantized neural network is implemented as a hardware module and integrated into the multi-legged robot for real-time detection. When a survivor is detected from robots, the corresponding information about GPS and the triangulation localization is thus delivered to the edge server. Then, rescuers can receive the notification message from the edge server by using their mobile devices. For survivor detection, experiments show the drone and the multilegged robot can achieve 2.164 fps and 2.404 fps, respectively.