LOCUS: A Multi-Sensor Lidar-Centric Solution for High-Precision Odometry and 3D Mapping in Real-Time

Achieving consistent lidar odometry in perceptually-degraded environments is a challenging problem. In these settings, self-similar areas, noisy sensor returns, high-rate rotations and large accelerations can greatly compromise the accuracy of the estimated odometry. While integrating additional sensor readings could help address these challenges, potential sensor failures must be robustly handled to not compromise the overall outcome of the mission. In this work, we present a lidar odometry system built for accurate and robust operation in extreme environments. The proposed system provides a failure-safe sensor integration module to flexibly include additional IMU and odometry measurements into lidar odometry. It is designed to be adaptable to different robotic platforms with diverse sensor inputs and computational capabilities to optimize real-time operation under challenging conditions. The system further provides a series of environment adaptation controls, including a flat ground assumption feature for operation in human-made buildings. We evaluate the performance of the proposed system against state-of-the-art techniques and discuss the enhancements achieved in localization accuracy and robustness to sensor failures. We also demonstrate real-time operation two different robots involved in exploration of the Satsop nuclear power plant, WA where the proposed system was a part of the CoSTAR team's solution that won first place at the Urban Circuit of the DARPA Subterranean Challenge.