Embodied Red Teaming for Auditing Robotic Foundation Models

TL;DR

This paper introduces Embodied Red Teaming (ERT), a novel evaluation method using automated techniques to generate challenging instructions for testing the safety and robustness of language-conditioned robotic models.

Contribution

The paper presents ERT, a new automated red teaming approach that creates diverse, challenging instructions grounded in context to evaluate robotic models beyond traditional benchmarks.

Findings

State-of-the-art models often fail on ERT instructions.

Models exhibit unsafe behaviors on challenging instructions.

Current benchmarks do not adequately assess safety or robustness.

Abstract

Language-conditioned robot models have the potential to enable robots to perform a wide range of tasks based on natural language instructions. However, assessing their safety and effectiveness remains challenging because it is difficult to test all the different ways a single task can be phrased. Current benchmarks have two key limitations: they rely on a limited set of human-generated instructions, missing many challenging cases, and focus only on task performance without assessing safety, such as avoiding damage. To address these gaps, we introduce Embodied Red Teaming (ERT), a new evaluation method that generates diverse and challenging instructions to test these models. ERT uses automated red teaming techniques with Vision Language Models (VLMs) to create contextually grounded, difficult instructions. Experimental results show that state-of-the-art language-conditioned robot models fail or behave unsafely on ERT-generated instructions, underscoring the shortcomings of current benchmarks in evaluating real-world performance and safety. Code and videos are available at: https://s-karnik.github.io/embodied-red-team-project-page.