OpenIoT & ELC Europe 2016

Indoor and outdoor navigation for small UAVs like quadrotors is still far from being a solved problem and most research and demonstration of visual navigation solutions have been conducted on traditional CPU architectures. Due to high computation demands of a full vision-based navigation stack (depth extraction, odometry, mapping, planning and control) a heavy and often power-hungry onboard computer is needed to simultaneously run all the tasks onboard.

For this presentation, we demonstrate a quadrotor UAV equipped with Nvidia's Jetson TX1 embedded GPU compute module and high-speed vision system, running our complete navigation stack onboard, at a tenth of the weight and power consumption. Our compact quadrotor is capable of tight, fast maneuvers and safe GPS-denied indoor hover, with onboard mapping, sense-and-avoid, precision landing and HD streaming.