Leg-KILO: Open-Sourcing Robust SLAM for Dynamic Legged Robots

rsasaki0109 · x · 2026-08-09

The Leg-KILO project has been open-sourced on GitHub, featuring a robust kinematic-inertial-lidar odometry system designed specifically for dynamic legged robots.

The project fuses LiDAR and IMU using a two-stage Error-State Kalman Filter (ESKF) frontend: Stage 1 performs incremental per-point updates along the scan timeline to compensate for motion distortion, while Stage 2 runs iterated ESKF (IESKF) over the full frame for global optimization. Additionally, the system maintains a hybrid feature Gaussian voxel map, improving feature utilization across both structured and unstructured environments.

Original post →

More from Embodied

Embodied channel →