RoboTrustBench: Benchmarking the Trustworthiness of Video World Models for Robotic Manipulation

2026-05-31

AI Summary

This work addresses the lack of systematic evaluation of trustworthiness in existing video world models for robotic manipulation, particularly in constraint-sensitive, counterfactual, and adversarial scenarios. We introduce RoboTrustBench, the first benchmark specifically designed to assess the trustworthiness of video world models in robotic tasks, built upon the real-world DROID dataset and encompassing four distinct scenario categories. Trustworthiness is formally defined and quantified across six dimensions with 13 fine-grained metrics, leveraging expert-validated instruction–image pairs, human evaluations, and scores from multimodal large language models. Experimental results reveal that while current models generate visually coherent videos, they exhibit significant shortcomings in constraint reasoning, counterfactual grounding, physical interaction modeling, and rejecting unsafe instructions.

Abstract

Video world models are increasingly used in robotic manipulation, yet existing benchmarks mostly evaluate them under valid, feasible, and safe instructions. We introduce RoboTrustBench, a benchmark for evaluating the trustworthiness of video world models under four scenarios: Normal, Constraint-Sensitive, Counterfactual, and Adversarial. Built from real-world DROID episodes, RoboTrustBench contains 1,207 expert-validated instruction-image pairs and a six-dimensional evaluation protocol with 13 fine-grained criteria. Evaluating seven representative video world models with human and MLLM assessment, we find that current models often generate visually coherent videos, but struggle with constraint reasoning, counterfactual grounding, physical interaction, and unsafe-instruction suppression. These results show that visual quality and surface-level instruction following are insufficient for trustworthy robotic video world modeling.

Problem

• video world models
• trustworthiness
• robotic manipulation
• benchmarking
• adversarial scenarios

Innovation

• video world models
• trustworthiness benchmark
• robotic manipulation
• counterfactual reasoning
• constraint-sensitive evaluation