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

2025-06-21

AI Summary

Deep-space robotic navigation in unknown, extreme terrains lacks formal safety guarantees—existing generative AI approaches prioritize performance over verifiable safety. Method: We propose a risk-guided dual-system coupled diffusion framework integrating a fast-learning “System 1” with a physics-driven, formally verifiable “System 2”. This is the first approach to enable computational sharing between training and inference phases, jointly ensuring generalization and formal safety. The framework unifies diffusion modeling, cognition-inspired architecture, real-time physics simulation, cross-modal robotic pretraining, and hardware-in-the-loop inference optimization. Contribution/Results: Evaluated on NASA JPL’s Mars analog terrain, our method reduces mission failure rate to 25% of baseline models while preserving target-reaching performance. Crucially, it achieves substantial robustness gains without additional training—demonstrating both theoretical soundness and practical efficacy for autonomous planetary exploration.

Technology Category

• Natural Language Processing: Safety and Robustness
• Intelligent Robots: Multi-Robot Systems
• Philosophy and Ethics of AI: Safety, Robustness & Trustworthiness

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

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 $4 imes$ while matching the goal-reaching performance of learning-based robotic models by leveraging inference-time compute without any additional training.

Problem

• Safe navigation in extreme unfamiliar terrain
• Limited safety guarantees in generative-AI methods
• Combining adaptability with formal safety

Innovation

• Risk-guided diffusion framework for safety
• Combines fast learned and slow physics-based systems
• Reduces failure rates without additional training