LiDAR-Visual-Inertial Fusion Architectures for GNSS-Resilient Aerial Navigation: Performance Assessment

This work proposes a performance assessment of LiDAR-Visual-Inertial navigation approaches specifically tailored for GNSS-denied or challenged circumstances. A data fusion scheme, based on Extended Kalman Filter, has been designed to face various environmental and flight conditions that occur depending on topology and morphological characteristics of the environment and UAV dynamic. It loosely integrates camera and LiDAR measurements. Their measurements are combined altogether exploiting LiDAR measurements to associate depth information with image features. Conversely, when camera measurements cannot be used, LiDAR odometry is performed to incrementally estimate the pose of the UAV by defining the pose transformation between one LiDAR point cloud and the following one, integrating the solution with the filter propagated state in order to bound the otherwise unbounded drift of IMU-only navigation solution. Sensors measurements are fed into the filter weighted by their own covariance, which indicates the level of confidence associated with the measurement. A set of gating criteria is introduced to assess the validity of each measurement to be used within the filter and thus to improve its resiliency against anomalous pose estimates. In general, particular attention is paid to the integrity of the multi-sensor-based solution. The strategy is assessed through numerical simulations conducted within the Unreal Engine 4 environment integrated with MATLAB and Simulink, which allows retrieving realistic sensors data by simulating aircraft trajectory, rich texture 3D geometry, and world-like scene. Results show percentage errors on the position estimates, computed with respect to the UAV traveled distance, below 0.9%, and errors on attitude angles estimate under 0.8°.