CDM: Contact Diffusion Model

A diffusion model for accurate multi-contact point localization using proprioceptive sensing.

← Back to Physical AI

A diffusion model for accurate multi-contact point localization using proprioceptive sensing.

Overview

CDM is a learning based method for locating multiple contacts on a robot from joint torque sensors and a force and torque sensor at the base. A diffusion model represents the multiple contact configurations that can produce the same sensor measurements, while signed distance field information helps the model reason over complex robot surfaces.

The method runs in real time and was validated in simulation and on a physical robot. It achieved localization errors of 0.44 cm for a single contact and 1.24 cm for two contacts in real world experiments.

Video

Publication

S. W. Han and M. J. Kim, “CDM: Contact Diffusion Model for Multi-Contact Point Localization,” IEEE ICRA, 2025. arXiv · IEEE