WCog-VLA: A Dual-Level World-Cognitive Vision-Language-Action Model for End-to-End Autonomous Driving

2026-07-09

AI Summary

This work addresses the limitation of existing end-to-end autonomous driving models, which lack holistic world understanding and are thus restricted to reactive driving behaviors. To enable proactive decision-making, the authors propose a dual-level world-aware Vision-Language-Action (VLA) framework that integrates semantic-level reasoning with generative-level evolutionary modeling. The core innovations include a novel semantic cognition mechanism combining 3D spatial perception, agent tokenization, and a game-theoretic chain-of-thought (Game-CoT), alongside an Aligned Decoupled Diffusion Transformer (ADDT) for efficient multi-agent trajectory generation. Evaluated on the NAVSIM benchmark, the method achieves state-of-the-art performance with a PDMS score of 92.9 and introduces a large-scale dataset comprising 85k Game-CoT annotations.

Abstract

Vision-Language-Action (VLA) models have advanced end-to-end autonomous driving. However, existing methods either lack comprehensive world cognition or suffer from fragmented world foresight, inherently confining these models to reactive driving. To address this limitation, we propose WCog-VLA, a novel dual-level World-Cognitive VLA framework that successfully bridges semantic world forecasting with generative world evolution to achieve proactive autonomous driving. At the semantic level, WCog-VLA unifies world cognition and reasoning by incorporating 3D spatial perception and injecting agent tokens to capture the world dynamics, while concurrently enabling Game-theoretic Chain-of-Thought (Game-CoT) reasoning. At the generative level, we introduce the Aligned Decoupled Diffusion Transformer (ADDT) as a powerful generative world model that synthesizes physically-plausible joint multi-agent trajectories. Through scene representation alignment, ADDT reduces the number of denoising steps required and thus significantly accelerates inference. To facilitate strategic reasoning, we further construct a large-scale dataset featuring 85k Game-CoT annotations. Extensive experiments on the NAVSIM benchmark demonstrate that WCog-VLA achieves a State-Of-The-Art (SOTA) PDMS score of 92.9.

Problem

• autonomous driving
• world cognition
• vision-language-action
• proactive driving
• world forecasting

Innovation

• World-Cognitive VLA
• Game-theoretic Chain-of-Thought
• Aligned Decoupled Diffusion Transformer
• Proactive Autonomous Driving
• Multi-agent Trajectory Generation