Back to feed
AR
arXiv CS.AI
6/24/2026
Neuro-Symbolic Drive: Rule-Grounded Faithful Reasoning for Driving VLAs

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

Short summary

Neuro-Symbolic Drive bridges symbolic AI planning and vision language model training by using rule-grounded reasoning traces from classical planners as supervision. Fine-tuning Qwen3.5-4B with these traces reduced driving error by 45% (ADE: 0.47→0.26) and collision miss-rate by 25% (8.30%→6.40%) in simulation. Code released on GitHub for practitioners.

  • •Pairs symbolic rule-based planner traces with VLM training to ensure reasoning is structurally coupled to motion generation
  • •Achieves 45% improvement in average displacement error and 25% reduction in miss rate on autonomous driving benchmarks
  • •Open-sourced implementation available; applicable to other embodied AI tasks requiring faithful reasoning

Generated with AI, which can make mistakes.

#research-breakthrough#ai-tools#ai-agents
Read full article at arXiv CS.AI

Is this a good recommendation for you?

Explore more

Human-like autonomy emerges from self-play and a pinch of human data
arXiv cs.LG
Think Twice, Act Once: Verifier-Guided Action Selection For Embodied Agents
arXiv CS.AI
PhoneHarness: Harnessing Phone-Use Agents through Mixed GUI, CLI, and Tool Actions
arXiv cs.CL
CoRA: Confidence-Rationale Alignment for Reliable Chain-of-Thought Reasoning
arXiv cs.CL