Run Your Visual-Inertial Odometry on NVIDIA Jetson: Benchmark Tests on a Micro Aerial Vehicle

This letter presents benchmark tests of various visual(-inertial) odometry algorithms on NVIDIA Jetson platforms. The compared include mono and stereo, covering Visual Odometry (VO) Visual-Inertial (VIO): VINS-Mono, VINS-Fusion, Kimera, ALVIO, Stereo-MSCKF, ORB-SLAM2 ROVIO. As these methods are mainly used for unmanned aerial vehicles (UAVs), they must perform well in situations where the size processing board weight is limited. boards released by satisfy constraints as have a sufficiently powerful central unit (CPU) graphics (GPU) image processing. However, existing studies, performance platform executing VO/VIO has not been extensively terms usage computing resources accuracy. Therefore, this study compares representative several platforms, namely TX2, Xavier NX, AGX Xavier, introduces novel dataset `KAIST VIO dataset' UAVs. Including pure rotations, geometric trajectories that harsh to state estimation. evaluation performed accuracy estimated odometry, CPU usage, memory boards, algorithms, trajectories. We present results comprehensive test release computer vision robotics applications.