Is the Future Compatible? Diagnosing Dynamic Consistency in World Action Models

2026-05-08

AI Summary

This work addresses a critical limitation in existing World Action Models (WAMs), whose generated trajectories rely solely on visual plausibility without verifying dynamic consistency between actions and state transitions, thereby compromising decision reliability. The study establishes action-state consistency as a core diagnostic criterion for WAM reliability, demonstrating its strong correlation with task success. It introduces a test-time multi-trajectory consensus selection mechanism that operates without value functions or explicit reward modeling. By leveraging joint forward prediction and inverse dynamics models to construct a consistency metric, the method significantly improves planning success rates on RoboCasa and RoboTwin 2.0. The results validate that action-state consistency is more indicative of decision quality than mere visual realism and further reveal that background collapse can adversely interfere with this consistency measure.

Abstract

World Action Models (WAMs) enable decision-making through imagined rollouts by predicting future observations and actions. However, the reliability of these imagined futures remains under-examined: is a generated future merely visually plausible, or is it dynamically compatible with the action sequence it claims to model? In this work, we identify action-state consistency, the alignment between predicted actions and induced state transitions, as a missing reliability axis for WAMs. Through a systematic study across representative joint-prediction and inverse-dynamics models, we find that action-state consistency systematically separates successful and failed rollouts across many tasks and follows similar success-failure trends as learned value estimates. These results suggest that consistency captures decision-relevant structure beyond visual realism. We further identify background collapse as an important boundary condition, where low-dynamics failed trajectories can become deceptively consistent because static futures are easier to predict. Building on these findings, we introduce a value-free consensus strategy for test-time selection, which ranks candidate rollouts by agreement among predicted futures. This strategy improves success rates on RoboCasa and RoboTwin 2.0 without additional training or reward modeling. Taken together, our findings establish action-state consistency as both a diagnostic tool for evaluating WAM reliability and a practical signal for value-free planning.

Problem

• World Action Models
• action-state consistency
• dynamic consistency
• imagined rollouts
• reliability

Innovation

• action-state consistency
• world action models
• dynamic compatibility
• value-free planning
• background collapse