Neuro-Symbolic Drive: Rule-Grounded Faithful Reasoning for Driving VLAs

2026-06-22

AI Summary

This work addresses the lack of causally grounded, structured semantic reasoning in existing vision-language-action (VLA) models for autonomous driving, which often leads to a disconnect between high-level decision-making and actual driving behavior. To bridge this gap, the authors propose a neuro-symbolic driving framework that, for the first time, converts the internal decision trajectories of classical rule-based planners into structured supervision signals to fine-tune the Qwen3.5-4B large language model. This approach integrates rule-based planning, multi-camera perception, Chain-of-Thought reasoning, and LLM fine-tuning to achieve constructive coupling between reasoning and action generation. Evaluated on both three-camera and eight-camera setups, the method significantly outperforms current baselines, reducing ADE@3s to 0.26 and miss rates to 6.40% and 5.99%, respectively.

Abstract

Driving VLA models incorporating Chain-of-Thought (CoT) reasoning are attractive because they leverage pretrained VLM representations and expose intermediate decisions in natural language, yet current rationales often lack the step-by-step decision semantics needed to keep the rationale causally connected to the planned motion. We introduce Neuro-Symbolic Drive, a neuro-symbolic driving framework that supervises a driving VLA with rule-grounded reasoning traces extracted directly from classical rule-based planners. Our key observation is that rule-based planners are symbolic AI systems that already function as executable reasoning engines: they reason about active safety constraints, search over candidate maneuvers, and select a final trajectory. We instrument these planners in simulation to capture both the executed trajectory and the internal decision trace at each rule-evaluation step. Each trace is serialized into structured rule-grounded reasoning and paired with the trajectory to fine-tune Qwen3.5-4B as a driving VLA. Because these traces are derived directly from the planner states that determine the action, they ensure reasoning is structurally coupled to motion generation by construction, rather than by post-hoc alignment. On our simulator-generated benchmark, detailed rule-grounded reasoning reduces ADE@3s from 0.47 to 0.26 and miss rate from 8.30% to 6.40% under three-camera perception, and from 0.54 to 0.26 and 10.13% to 5.99% under eight-camera perception. Neuro-Symbolic Drive thus converts neuro-symbolic planning logic into structured supervision. Code base: https://github.com/XiangboGaoBarry/Neural-Symbolic-Drive.

Problem

• Chain-of-Thought reasoning
• visual-language-action models
• causal reasoning
• driving decision semantics
• motion planning

Innovation

• Neuro-Symbolic
• Rule-Grounded Reasoning
• Driving VLA
• Chain-of-Thought
• Executable Planning