ChainFlow-VLA: Causal Flow Planning with Vision-Language Models

2026-05-22

AI Summary

This work addresses a fundamental mismatch in existing end-to-end autonomous driving systems between temporal causal reasoning and global trajectory consistency: autoregressive models suffer from error accumulation, while diffusion models lack explicit causal constraints. To reconcile these limitations, the authors propose a hybrid probabilistic framework that first generates causally structured trajectory modes via an autoregressive module and then leverages the hidden states of a vision-language model (VLM) as semantic priors to guide a diffusion-based refiner in residual space for fine-grained, semantics-aware global optimization. This approach uniquely integrates autoregressive causal modeling with diffusion-based global refinement, achieving a score of 94.85 on the NAVSIM v1 benchmark—matching human-level performance (94.8)—and significantly enhancing planning robustness in ambiguous and long-tail scenarios.

Abstract

Current end-to-end autonomous driving systems are fundamentally limited by a mismatch between temporal causal reasoning and global trajectory consistency. Autoregressive (AR) models capture interaction-aware temporal dependencies via causal factorization, but their step-wise decoding leads to error accumulation and suboptimal global structure. In contrast, diffusion models optimize trajectories globally but lack explicit causal constraints, making them unreliable in interactive and safety-critical scenarios. This dichotomy reveals a deeper issue: existing methods treat causal modeling and global optimization as separate paradigms, without a principled way to unify them within a single trajectory distribution. To address this, we propose ChainFlow-VLA, which unifies causal generation and global refinement within a unified probabilistic framework. We formulate planning as a mixture over AR-induced modes and learn Vision-Language Model (VLM)-conditioned residual distributions over these modes. An autoregressive generator (Chain) produces a discrete set of causal trajectory modes, followed by a diffusion-based refiner (Flow) that leverages VLM hidden states as semantic priors to perform mode-conditioned correction in residual space while preserving causal structure. This straightforward conditioning seamlessly injects high-level scene understanding into fine-grained trajectory adjustments. Experiments demonstrate that ChainFlow-VLA achieves robust planning in ambiguous and long-tail scenarios, achieving a state-of-the-art score of 94.85 on the NAVSIM v1 leaderboard, matching human-level performance (94.8). Code will be available at https://github.com/AFARI-Research/ChainFlow-VLA.

Problem

• autonomous driving
• causal reasoning
• trajectory planning
• global consistency
• vision-language models

Innovation

• causal planning
• diffusion refinement
• vision-language model
• autoregressive trajectory generation
• residual space optimization