Verifiable Foundation Models for Robot Safety

2026-06-21

AI Summary

This work addresses the challenge of applying highly expressive foundation models to robot control in safety-critical settings, where their inherent complexity hinders formal verification. To reconcile expressiveness with verifiability, the authors propose the FEARL framework, which decomposes the policy into a large-model controller (C) and a compact safety module (S). Controller C handles multimodal perception and high-level task reasoning, while safety module S—receiving only low-dimensional inputs from dedicated safety sensors—executes actions and is amenable to formal verification. This modular design enables, for the first time, foundation model–driven policies that retain high task performance while supporting rigorous safety guarantees. The approach also facilitates sim-to-real transfer. Experiments across three simulated tasks and on a physical robot demonstrate that FEARL effectively balances safety assurance with task effectiveness.

Abstract

Deploying foundation models for robot control raises a central challenge: the expressive power that enables rich, multimodal perception also makes these models opaque and difficult to analyze formally, rendering them intractable for existing verification tools. In this paper, we present FEARL (Foundation-Enabled Assured Robot Learning), a framework that addresses this tension through a modular architectural decomposition. FEARL separates the policy into a large Controller (C) responsible for high-dimensional perception and task reasoning, and a small Safety module (S) that receives low-dimensional observations from dedicated safety sensors together with a bounded context embedding from C and produces the final action. Since many robot safety requirements, such as collision avoidance and workspace boundary constraints, can be expressed over these safety sensor observations, formal verification can be applied to S rather than to the full foundation-model backbone. This makes formal analysis tractable with existing tools while preserving the Controller's expressive power for task reasoning. To show that the decomposed policy remains capable of solving diverse tasks, we evaluate FEARL on three simulated robotic domains using multiple Controller backbones and training procedures, including pretrained off-the-shelf vision-language-action models. We further transfer the learned policy from one of our simulated tasks to a physical robot, suggesting that the low-dimensional safety interface supports practical sim-to-real transfer.

Problem

• foundation models
• robot safety
• formal verification
• verifiability
• safety assurance

Innovation

• verifiable AI
• foundation models
• robot safety
• modular architecture
• formal verification