FootRecon: Quadrupedal Terrain Reconstruction from Sparse Foot Contacts with Geometric Prior

Yujin Park, Seungoh Han, Taebok Lee, CHANWOO LEE, Hyungyu Park, Kyungdon Joo

Keywords: Legged Robot Perception, Proprioceptive Terrain Reconstruction

TL;DR: proprioceptive-only terrain reconstruction by resolving sparse-contact ambiguity with a learned generative prior and geometry-aware optimization

Abstract: Reliable terrain geometry is essential for safe legged locomotion in unstructured environments. However, existing methods rely on dense exteroceptive sensing, whose reliability degrades under adverse conditions. In contrast, proprioceptive foot-terrain contacts remain available but are inherently sparse, rendering terrain reconstruction fundamentally ill-posed. We propose FootRecon, a proprioceptive-only terrain reconstruction framework that formulates geometry estimation as a contact-conditioned generative inference problem. A conditional variational autoencoder (cVAE) learns structural terrain priors to resolve ambiguity in underconstrained regions, while a geometry-aware optimization enforces contact consistency and preserves discontinuities. The refined local patches are incrementally fused into a globally consistent 2.5D height map during locomotion. Real-world experiments across diverse outdoor terrains demonstrate improved geometric fidelity over contact-only baselines while maintaining online performance.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Paper Acceptance: Yes

Submission Number: 26