Field: Computer Science > Artificial Intelligence
arXiv ID: 2606.23938 (cs)
Submitted: 22 June 2026
Authors: Xiangbo Gao, Xiukun Huang, Boyu Lu, Junge Zhang, Mengjie Mao, Jiachen Li, Wei Xiong, Zhengzhong Tu
Abstract
Driving Vision-Language-Action (VLA) models that incorporate Chain-of-Thought (CoT) reasoning are increasingly attractive because they leverage pretrained VLM representations and expose intermediate decisions as natural language. However, existing rationales often lack the step-by-step decision semantics needed to maintain a causal connection between the rationale and the planned motion. To address this, 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 corresponding 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 relying on post-hoc alignment. On our simulator-generated benchmark, the proposed rule-grounded reasoning approach yields substantial improvements:
- Under three-camera perception: Average Displacement Error at 3 seconds (ADE@3s) reduces from 0.47 to 0.26, and miss rate from 8.30% to 6.40%.
- Under eight-camera perception: ADE@3s drops from 0.54 to 0.26, and miss rate from 10.13% to 5.99%.
Neuro-Symbolic Drive thus converts neuro-symbolic planning logic into structured supervision, enabling more faithful and causally grounded decision-making in autonomous driving systems.
Code is available at: https://github.com/XiangboGaoBarry/Neural-Symbolic-Drive
Subjects: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
Cite as: arXiv:2606.23938 [cs.AI] (or arXiv:2606.23938v1 [cs.AI] for this version)
via ArXiv AI