WorldGuide improves vision language action policies by learning failure boundary
WorldGuide: Learning Success-Failure Boundaries in Latent World Models for Vision-Language-Action Policies
Computer Vision and Pattern Recognition
Summary
Vision-language-action systems help robots or AI understand and act based on what they see and hear. Traditional training mostly teaches these systems what success looks like, so they struggle to tell when an action fails, especially when failures look similar to successes. The authors propose WorldGuide, which trains a model to recognize the difference between success and failure in a hidden representation space. This helps guide the AI to learn better policies without needing extra model computations after training.
What this means in practice
- •For robotics developers: Improve robot control systems to better distinguish between successful and failed actions using visual and language inputs in training.
- •For game environment designers: Enhance AI agents in simulated environments to reliably learn from both positive and negative outcomes during training.
Authors
Lin Liu, Lu Zhang, Ziying Song, Wu Yang, Yuzheng Zhuang, Yunzhi Zhuge, Shuai Tao, Wulong Liu, Huchuan Lu
Abstract
Latent world models offer a promising way to improve Vision-Language-Action policies by capturing the consequences of actions. However, models trained primarily on expert demonstrations have limited exposure to failure outcomes and may struggle to distinguish visually similar successful and failed interactions. We propose \textbf{WorldGuide}, a framework that learns these distinctions in latent space and uses them to guide policy training. WorldGuide combines predictive pretraining on successful and failed trajectories with contrastive learning on matched success--failure pairs. The learned predictor then provides a differentiable reward to guide joint optimization of the policy and visual encoder. The predictor is discarded after training, so deployment requires no additional world-model inference. Extensive experiments show that WorldGuide substantially improves VLA reliability and achieves state of the art performance on LIBERO 100 and SimplerEnv, reaching \textbf{96.8\%} and \textbf{72.0\%}, respectively. Code will be publicly available.