Predict the Physics-Informed Terrain Properties Over Deformable Soils using Sensorized Foot for Quadruped Robots

Abstract

Unstructured environments pose diverse terrain-related risks, including slip and sinkage, that could significantly impact legged robot locomotion. Avoiding these risks can be achieved by predicting terrain properties using a physical foot-terrain interaction model. Advanced foot design based on this model, instead of traditional empirical models, can lead to an improved overall traction performance. However, this still represents an open research problem. In this paper, we propose a planar foot design with a tactile-based perception that can adapt to deformable terrains while estimating multi-modal contact states and relative terrain contact processes. It also uses a unified foot-terrain interaction model that informs contact terrain properties over the contact region. Furthermore, we use this pipeline to improve foot sole design. We validate its performance and compare multiple designs using both single- foot and quadruped robot locomotion experiments. The results indicate that our proposed method can reduce terrain-related risks over different soft terrains.