RoboWorld: Fast and Reliable Neural Simulators for Generalist Robot Policy Evaluation

2026-07-01

AI Summary

Existing video world models suffer from unreliable trajectory generation and low inference efficiency when evaluating general-purpose robotic policies, hindering their scalability. To address this, this work proposes an efficient and reliable automated evaluation pipeline that integrates a Step Forcing training strategy, which combines anchored and single-step autoregressive contexts to mitigate train-test mismatch while preserving action-observation dynamics consistency, thereby significantly enhancing the reliability of long-horizon simulation. Coupled with a fast autoregressive video world model and a task-progress-aware vision-language model scoring mechanism, the framework enables accurate policy performance assessment. Experiments demonstrate strong alignment with real-world robot evaluations across multiple tasks and environments, achieving Pearson and Spearman correlation coefficients of 0.989 and 0.970, respectively.

Abstract

Video world models are emerging as a scalable alternative for evaluating generalist robot policies, bypassing the physical constraints and engineering burdens of real-world deployment. However, evaluating policies with video world models remains challenging, as world-model errors can make generated rollouts unreliable and slow inference limits large-scale throughput. We introduce RoboWorld, an automated evaluation pipeline that pairs a fast autoregressive video world model with a task-progress-aware vision-language model scoring. To enable reliable long-horizon autoregressive world-model rollouts, we propose Step Forcing, which combines anchored and one-step self-forwarded contexts to reduce train--test mismatch while preserving action--observation dynamics. Together, these components enable RoboWorld to align strongly with real-world robot evaluation across tasks and environments, achieving Pearson's r = 0.989 and Spearman's \r{ho} = 0.970.

Problem

• video world models
• robot policy evaluation
• world-model errors
• slow inference
• long-horizon rollouts

Innovation

• video world models
• Step Forcing
• vision-language scoring
• robot policy evaluation
• autoregressive simulation