KING: Embodiment-Aware Kinematic Graph Neural Network for Unified Motion Representation of Legged and Wheeled Robots

2026-08-02

AI Summary

Existing learning-based kinematic models struggle to generalize to novel robot morphologies due to structural and semantic discrepancies in proprioceptive signals. This work proposes a unified kinematic modeling approach based on graph neural networks, which, for the first time, encodes both wheeled and legged robot configurations into a common graph representation, enabling a single model to achieve high-accuracy odometry estimation across diverse robot types. Requiring only a URDF file and minimal data—approximately one minute of real-world interaction—the method rapidly adapts to new morphologies without retraining from scratch. By fusing proprioceptive and IMU measurements and leveraging few-shot transfer learning, the approach substantially outperforms conventional methods, achieving notable advances in both generalization capability and estimation accuracy.

Abstract

Kinematic models provide reliable motion constraints for odometry estimation in featureless environments, where exteroceptive sensing degrades and IMU integration drifts. Learning-based kinematic models can achieve more accurate odometry estimation than model-based methods by capturing nonlinear effects; however, most existing learning-based models are trained on a single embodiment and generalize poorly to new embodiments. This generalization is difficult because the meanings and structures of proprioceptive measurements vary across embodiments, including the number of joints and ground-contact elements (e.g., wheels, feet). To address this challenge, we propose KING, a Graph Neural Network (GNN)-based kinematic model that explicitly incorporates robot embodiments by representing them as a common graph. We show that wheel and leg kinematic models can be expressed by a unified representation, enabling a single model for both wheeled and legged robots. Trained on datasets spanning diverse embodiments, KING provides a unified representation of wheeled and legged kinematics and achieves high-accuracy odometry estimation in real environments. KING estimates accurate odometry using only an embodiment description (e.g., a URDF file) and on-board proprioception (encoders and an IMU) and can be adapted to new robot embodiments through few-shot learning with only one minute of data, avoiding retraining from scratch on a new dataset for each robot. The project page is available at: https://smrg-aist.github.io/king_project_page/

Problem

• kinematic modeling
• embodiment generalization
• legged robots
• wheeled robots
• odometry estimation

Innovation

• Graph Neural Network
• Unified Kinematic Representation
• Embodiment-Aware Learning
• Few-Shot Adaptation
• Legged and Wheeled Robots