Wildcat: Online Continuous-Time 3D Lidar-Inertial SLAM

Abstract

We present Wildcat, a novel online 3D lidar-inertial SLAM system with exceptional versatility and robustness. At its core, Wildcat combines a robust real-time lidar-inertial odometry module, utilising a continuous-time trajectory representation, with an efficient pose-graph optimisation module that seamlessly supports both the single-and multi-agent settings. The robustness of Wildcat was recently demonstrated in the DARPA Subterranean Challenge where it outperformed other SLAM systems across various types of sensing-degraded and perceptually challenging environments. In this paper, we extensively evaluate Wildcat in a diverse set of new and publicly available real-world datasets and showcase its superior robustness and versatility over two existing state-of-the-art lidar-inertial SLAM systems.

Hilti-Oxford Dataset: A Millimeter-Accurate Benchmark for Simultaneous Localization and Mapping

“…The highest scoring team was CSIRO with a score of 563.8. Their Wildcat SLAM [15] algorithm uses a continuous-time trajectory representation for lidar-inertial odometry using slidingwindow optimization and online pose graph optimization. This is refined by an offline global optimization module that takes advantage of non-causal information.…”

Quatro++: Robust global registration exploiting ground segmentation for loop closing in LiDAR SLAM

“…Most LiDAR SLAM methods (Li et al, 2021;Ramezani et al, 2022;Shan et al, 2020;Shan and Englot, 2018) heavily rely on mean squared error (MSE) of local registration methods to determine whether the loop candidate is considered as a false loop or not, that is the output of getFitnessScore() in point cloud library (Aldoma et al, 2012;Rusu and Cousins, 2011). The MSE is the average of squared Euclidean distances from the transformed source cloud points by the estimated relative rotation and translation to the corresponding target cloud points (Rusu and Cousins, 2011).…”

LiDAR Ground Segmentation and Modeling for Mobile Robots in Unstructured Terrain

“…The main advantages of LiDAR inertial odometry, enabled by IMU measurements, are higher robustness in fast motions, stabilization in feature poor locations and the observability of the gravitational direction to stabilize the drift of orientation. Wildcat SLAM [17] extracts surfels from the point clouds and optimizes a continuous trajectory defined by control poses and spline interpolation in a sliding window manner together with the IMU data over several seconds to obtain highly accurate submaps. For each submap, a gravitation direction is determined and submaps are registered relative to each other.…”

“…Most of the LiDAR (inertial) odometry methods mentioned above build incremental maps the current scan is aligned to. While some map the environment without optimizing the consistency within the map [20], others perform featurebased optimization only in a sliding time window [12], [17]. After loop closure detections, a pose graph optimization is usually performed.…”

“…2) Sliding Window DMSA Continuous Trajectory Optimization: In this step, a continuous trajectory of a sliding time window of duration d window is optimized. The modeling of the trajectory was inspired by [17]. The trajectory is represented by a set of control poses uniformly distributed in time.…”