The article evaluates local navigation systems for mobile robots, specifically focusing on visual and lidar odometry. The author tests several algorithms, including ORB-SLAM 3, RtabMap, OpenVins, Fast-LIO 2, and DLIO, to determine their reliability in outdoor environments without external infrastructure.
ORB-SLAM 3 performance was tested in various conditions. In environments with rich geometry, the system functioned well, but it struggled with drift and tracking loss when moving through open areas with poor geometric features or when clouds were incorrectly identified as reference points. The system frequently required manual recalibration after tracking failures.
The author noted that the RGB-D plus Inertial mode outperformed the standard RGB-D mode. However, the algorithm still failed to provide high-precision positioning, showing significant drift and unpredictable crashes, making it unsuitable for reliable outdoor robot odometry.
Other tested packages showed varying results. Stella vslam and REVO failed due to sensitivity to image jitter. REBiVO and VI Stereo DSO suffered from low performance on an i7 mini-PC, failing to operate in real-time. Vins mono struggled with IMU synchronization issues when used with a RealSense D455 sensor.
