New SLAM method cuts elevation drift from 30m to 30cm using leg odometry
Legged robots now navigate GNSS-denied areas with 100x better vertical accuracy
A team of robotics researchers (Léon Perruchot-Triboulet, Luc Jaulin, Kai Xiao) has published a paper on arXiv demonstrating a novel approach to SLAM (Simultaneous Localization and Mapping) for legged robots operating in GNSS-denied environments. Their method augments the popular LIO-SAM (LiDAR-Inertial Odometry and Mapping) framework with a parallel kinematic lane driven by proprioceptive leg odometry. This additional lane is coupled to the main LiDAR-inertial lane via an identity relative pose constraint with a selective noise model, essentially using the robot's own leg movement data as a vertical anchor.
Tested on a Linxai D50 quadruped robot across two outdoor loops totaling over one kilometer, the results are striking: elevation drift dropped from over 30 meters (which makes traditional SLAM unusable for navigation) to under 30 centimeters — a 100x improvement. In one scene where the baseline LIO-SAM pipeline failed entirely (unable to converge), the new method successfully completed the trajectory. Because the proprioceptive data is already computed onboard for gait control, this enhancement is lightweight and requires no additional sensors. The work, submitted to the ICRA workshop on Robot Meets GNSS and Ranging for Seamless Autonomy, highlights how internal robot cues can solve a critical weakness of LiDAR-based SLAM in geometrically sparse or repetitive environments.
- Reduces elevation drift from >30m to <30cm on a Linxai D50 quadruped over 1km+ outdoor loops
- Augments LIO-SAM with a parallel leg odometry lane using existing proprioceptive data from gait control
- Enables convergence in scenes where baseline LIO-SAM fails entirely, at zero hardware cost
Why It Matters
Enables reliable autonomous navigation for legged robots in GPS-denied areas like mines, forests, and indoor sites.