Resilient pose estimation for autonomous systems, and especially small unmanned aerial robots, is one of the core capabilities required for these robots to perform their assigned tasks in a reliable and efficient manner. Different sensing modalities have been utilized for the robot pose estimation process, particularly in GPS–denied environments. However, as aerial robots are deployed in more complex environments, such as subterranean mines and tunnels, different sensing modalities can become degraded in different parts of the environment due to the diversity of sensor perception challenges presented in terms of both nature and condition of the operational environment. Motivated by this fact, in this work a complementary multi–modal sensor fusion approach is presented that improves the reliability of the pose estimation process for aerial robots by fusing visual–inertial (VIO) and thermal–inertial (TIO) odometry estimates with a LiDAR odometry and mapping solution. In particular, VIO/TIO estimates are utilized for providing robust priors for LiDAR pose estimation as well as for selectively propagating the LiDAR pose estimates when LiDAR pose estimation process becomes degenerate. The proposed approach is experimentally verified in a variety of subterranean environments as well as utilized during the competition run of the tunnel circuit of the DARPA Subterranean Challenge.
Complementary Multi–Modal Sensor Fusion for Resilient Robot Pose Estimation in Subterranean Environments
2020-09-01
2678297 byte
Conference paper
Electronic Resource
English
Multi-Modal Neural Feature Fusion for Pose Estimation and Scene Perception of Intelligent Vehicle
SAE Technical Papers | 2021
|Multi-Modal Neural Feature Fusion for Pose Estimation and Scene Perception of Intelligent Vehicle
British Library Conference Proceedings | 2021
|Multi‐Sensor Geometric Pose Estimation
Wiley | 2022
|