Risk-Guided Diffusion: Toward Deploying Robot Foundation Models in Space, Where Failure Is Not An Option

TL;DR

This paper introduces a risk-guided diffusion framework for space robot navigation that combines learned and physics-based systems, significantly reducing failure rates while maintaining goal-reaching performance.

Contribution

It proposes a novel fusion of fast learned and slow physics-based systems for safe, reliable robot navigation in space environments, inspired by human cognition.

Findings

Reduces failure rates by up to 4 times in Mars-analog experiments.

Maintains goal-reaching performance comparable to existing learning models.

Leverages inference-time compute without additional training.

Abstract

Safe, reliable navigation in extreme, unfamiliar terrain is required for future robotic space exploration missions. Recent generative-AI methods learn semantically aware navigation policies from large, cross-embodiment datasets, but offer limited safety guarantees. Inspired by human cognitive science, we propose a risk-guided diffusion framework that fuses a fast, learned "System-1" with a slow, physics-based "System-2", sharing computation at both training and inference to couple adaptability with formal safety. Hardware experiments conducted at the NASA JPL's Mars-analog facility, Mars Yard, show that our approach reduces failure rates by up to $4\times$ while matching the goal-reaching performance of learning-based robotic models by leveraging inference-time compute without any additional training.