LILO: Light Detection and Ranging, Inertial and Leg Odometry Combined Simultaneous Localization and Mapping Based on Kalman Filter for Legged Robots
Abstract
The Simultaneous Localization and Mapping (SLAM) performance of a mobile robot is affected by many factors, including the algorithm, sensor scheme and mobile platform, etc. Compared with wheeled robots, the SLAM performance of legged robots is weak with the same algorithm and sensor configurations since the unstable legged locomotion leads to the body vibration, which is unfavorable to the data sampling of sensors. To address this issue, we propose an improved LiDAR-Inertial-Leg Odometry (LILO) algorithm that tightly couples light detection and ranging (LiDAR), inertial measurement unit (IMU), and leg odometry pre-integration factors within a factor graph framework. The algorithm leverages IMU and leg odometry information for drift-free motion priors while LiDAR provides structural constraints, and all measurements are fused through Kalman filtering and nonlinear optimization in Georgia Tech Smoothing and Mapping (GTSAM). The Laser SLAM and visual SLAM with different algorithm are compared on wheeled and legged robots, and comparative experiments demonstrate that the proposed LILO framework effectively mitigates the impact of vibrational noise, enabling a legged robot to achieve reliable SLAM with an overall mapping accuracy of 4.6% mean relative absolute error in challenging environments, significantly outperforming standard single-sensor and leg-agnostic fusion approaches.