Safe Active Navigation and Exploration for Planetary Environments Using Proprioceptive Measurements

TL;DR

Problem: Vision-based planning struggles on deformable granular terrains in planetary settings, motivating a proprioception-driven approach. Method: SAEGT fuses Gaussian Process-based traversability mapping from per-step contacts with online safe and frontier region detection and a reactive geometry-based controller for real-time safe exploration. Key contributions: online traversability estimation with uncertainty, Lipschitz-based safe region expansion, and frontier-driven exploration that can progress toward goals or map unknown terrain. Findings: in simulation, the framework safely navigates around hazardous zones and can perform goal-free exploration using only proprioceptive measurements, enabling effective planetary scouting under limited sensing.

Abstract

Legged robots can sense terrain through force interactions during locomotion, offering more reliable traversability estimates than remote sensing and serving as scouts for guiding wheeled rovers in challenging environments. However, even legged scouts face challenges when traversing highly deformable or unstable terrain. We present Safe Active Exploration for Granular Terrain (SAEGT), a navigation framework that enables legged robots to safely explore unknown granular environments using proprioceptive sensing, particularly where visual input fails to capture terrain deformability. SAEGT estimates the safe region and frontier region online from leg-terrain interactions using Gaussian Process regression for traversability assessment, with a reactive controller for real-time safe exploration and navigation. SAEGT demonstrated its ability to safely explore and navigate toward a specified goal using only proprioceptively estimated traversability in simulation.

Figures (5)

• Figure 1: Geometric processing pipeline for converting safe terrain points to navigable free space representations. (a) Visualization of a concave hull generated from point clusters. Image adapted from cubao2024concavehull. Licensed under the MIT License. (b) Bounding box creation for workspace definition.
• Figure 2: Ground truth traversability values for experiments.
• Figure 3: A single frame showing the navigation regions and setup. White = safe terrain ($f(\mathbf{x}) \geq h$); Gray = low-traversability terrain ($f(\mathbf{x}) < h$); Black = obstacles ($\mathcal{O}_t$); Green = local freespace; Light Blue = safe region ($S_t$); Red = frontier region ($G_t$); Purple = intermediate goal; Dark Blue = final goal.
• Figure 4: Navigation around low-traversability zone ($h = 1000$)
• Figure 5: Overall exploration progression over 9 intervals.