Single-shot Foothold Selection and Constraint Evaluation for Quadruped Locomotion

TL;DR

This paper introduces a neural network-based method for rapid, single-step foothold selection and constraint evaluation for quadruped robots navigating rough terrain, enabling efficient real-time decision-making.

Contribution

It presents a novel CNN approach that evaluates multiple footholds and constraints simultaneously in 10 ms, improving speed and reliability over traditional methods.

Findings

Fast foothold evaluation within 10 ms on standard hardware

Effective navigation over rough terrain in simulation and real-world tests

Neural network accurately assesses geometrical and kinematic constraints

Abstract

In this paper, we propose a method for selecting the optimal footholds for legged systems. The goal of the proposed method is to find the best foothold for the swing leg on a local elevation map. We apply the Convolutional Neural Network to learn the relationship between the local elevation map and the quality of potential footholds. The proposed network evaluates the geometrical characteristics of each cell on the elevation map, checks kinematic constraints and collisions. During execution time, the controller obtains the qualitative measurement of each potential foothold from the neural model. This method allows to evaluate hundreds of potential footholds and check multiple constraints in a single step which takes 10~ms on a standard computer without GPGPU. The experiments were carried out on a quadruped robot walking over rough terrain in both simulation and real robotic platforms.