Tightly-Coupled LiDAR-IMU-Leg Odometry with Online Learned Leg Kinematics Incorporating Foot Tactile Information
Masashi Yokozuka, Kentaro Uno, and Kazuya Yoshida
AI summary
Problem
Conventional legged robot odometry fails in featureless environments due to LiDAR degradation and on deformable terrain due to foot slippage, while IMU-based methods suffer from rapid drift. Existing approaches also lack adaptability to dynamic changes in robot weight load and terrain conditions.
Approach
The authors fuse LiDAR, IMU, and leg kinematics in a tightly coupled factor graph, using a neural network that learns foot-ground dynamics online from tactile sensor data. This unified framework jointly optimizes odometry and network parameters while estimating constraint uncertainty in real time.
Key results
- Neural leg kinematics model incorporating foot tactile information for adaptive motion prediction
- Unified factor graph framework jointly optimizing odometry and online network training
- Online uncertainty estimation for leg kinematics constraints to handle deformable terrain
- Real-world quadruped experiments showing superior accuracy over state-of-the-art methods in featureless areas and terrain changes
Why it matters
Enables reliable navigation for delivery and inspection legged robots in complex, real-world environments where traditional sensors fail.
Abstract
In this letter, we present tightly coupled LiDAR- IMU-leg odometry, which is robust to challenging conditions such as featureless environments and deformable terrains. We developed an online learning-based leg kinematics model named the neural leg kinematics model, which incorporates tactile infor- mation (foot reaction force) to implicitly express the nonlinear dynamics between robot feet and the ground. Online training of this model enhances its adaptability to weight load changes of a robot (e.g., assuming delivery or transportation tasks) and terrain conditions. According to the neural adaptive leg odometry factor and online uncertainty estimation of the leg kinematics model-based motion predictions, we jointly solve online training of this kinematics model and odometry estimation on a unified factor graph to retain the consistency of both. The proposed method was verified through real experiments using a quadruped robot in two challenging situations: 1) a sandy beach, representing an extremely featureless area with a deformable terrain, and 2) a campus, including multiple featureless areas and terrain types of asphalt, gravel (deformable terrain), and grass. Experimental results showed that our odometry estimation incorporating the neural leg kinematics model outperforms state- of-the-art works. Our project page is available for further details: https://takuokawara.github.io/RAL2025 project page/