Learning Contact Representation for Leg Odometry
Self-supervised foot-contact estimation from joint encoders and IMU only, fed as a stance belief into a Kalman filter. Reports the lowest leg-odometry error among the compared baselines.仅用关节编码器与 IMU 自监督估计足端接触，并作为支撑置信度输入卡尔曼滤波器。报告腿式里程计误差在对比方法中最低。
BibTeX
@misc{girgin2026learning,
title = {{Learning Contact Representation for Leg Odometry}},
author = {Emre Girgin and Cagri Kilic},
year = {2026},
eprint = {2606.05501},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.05501},
}Locomotion & Humanoids运动控制与人形机器人Perception, SLAM & 3D Mapping感知、SLAM 与三维建图leggedstate-estimationcontact-detectionself-supervisedkalman-filter