Working Draft · Initiative Phase

Technical Working Group for Formal Automotive Systems

Building an open, vendor-neutral standard for formal behavioral representation in AI-driven automotive systems — enabling transparent, verifiable and certifiable autonomous driving technology.

Join the Initiative Learn More

Vehicle Machine Language (VML)

VML is envisioned as an open standard that serves as the formal behavioral representation inside future AI-based driving systems.

Instead of allowing natural language to directly influence runtime decision making, VML introduces a formally defined intermediate representation that can be interpreted, analyzed, verified and executed with deterministic semantics.

In this architecture, foundation models such as LLMs and Vision-Language Models become semantic compilers that translate human requirements into formal specifications. Runtime systems operate exclusively on VML.

Natural language belongs to engineering.
Formal languages belong to runtime systems.

This separation provides a clear distinction between human communication and machine execution, while preserving the full benefits of modern AI-assisted engineering.

The long-term vision is a transformation chain:
Natural Language → VML → Vehicle Behavior

The VML Pipeline

From human requirements through a foundation model as semantic compiler, through VML as the formal runtime representation, to formal verification, trajectory generation and vehicle execution.

VML Architecture Diagram showing pipeline from Human Requirements through Foundation Model, VML, Formal Verification, Planner/Optimizer to Vehicle Controller, with VML Language Layers and Runtime Monitoring. click to enlarge

Figure 1 — The VML pipeline: engineering-time natural language is compiled into formal VML by a foundation model; runtime components operate exclusively on VML.

Why Formal Representation?

Modern automotive AI systems increasingly combine perception, reasoning and planning into unified Vision-Language-Action architectures. While these systems demonstrate impressive capabilities, natural language was never designed as an execution language for safety-critical systems.

Natural language is inherently ambiguous, context-dependent and open to interpretation — properties that make it excellent for human communication but difficult to verify, certify and validate as part of an automotive runtime system.

The Problem

Natural language ambiguity at runtime creates non-deterministic behavior that cannot be systematically verified or certified.

🔬

The Principle

A formal intermediate representation separates engineering communication from machine execution, making each independently optimizable.

The Outcome

Automotive AI becomes more transparent, testable and certifiable without restricting future advances in foundation models.

Vendor-Neutral, Open, Collaborative

Future automotive AI should not depend on proprietary behavioral languages. An open standard allows the entire automotive ecosystem to build on a common foundation.

The goal is not to standardize AI models themselves, but to standardize the formal representation of behavioral intent that connects engineering, verification and runtime execution.

An open standard enables

  • Interoperability between tools
  • Independent verification
  • Vendor-neutral implementations
  • Transparent certification
  • Long-term maintainability
  • Academic research
  • Regulatory collaboration
  • Broad industrial adoption

What VML Delivers

🛡

Functional Safety

Deterministic semantics suitable for analysis and formal verification under ISO 26262 and related standards.

📋

Regulatory Transparency

Behavioral rules can be disclosed to authorities without revealing proprietary AI implementations.

🔍

Formal Verification

Behavior becomes amenable to model checking, theorem proving, runtime monitoring and systematic testing.

💡

Explainability

Decisions trace back to explicit formal rules instead of opaque natural-language prompts.

Better Development

VML provides a common interface between requirements engineering, AI-assisted specification, simulation, testing and deployment.

🔄

Technology Independence

The formal language remains independent of perception algorithms, planning methods and future AI architectures.

Topics Under Development

  • Vehicle Machine Language (VML)
  • Formal behavioral semantics
  • Runtime execution models
  • Temporal & deontic reasoning
  • Probabilistic behavior specification
  • Formal ontologies
  • Verification interfaces
  • Runtime monitoring
  • Testing methodologies
  • Reference implementations
  • Conformance testing

Specifications & Working Documents

The following documents represent the current working drafts. All specifications are developed under an open process and will be published as open standards.

Join the Initiative

The initiative is currently in its initial working draft phase with one founding member. We are actively seeking contributors to establish a consortium for developing this open standard.

All technical discussions, working drafts and future specifications are developed through an open and collaborative process. Contributions from any organization or individual aligned with the vision are welcome.

Automotive OEMs Tier-1 Suppliers Semiconductor Vendors Software Companies Universities Research Institutes Standardization Organizations Government Agencies Independent Researchers
Get in Touch

Get Involved

If you are interested in contributing to the working group, have questions about the specifications, or would like to explore collaboration opportunities, please reach out.

By submitting this form you agree to be contacted regarding TWG·FAS activities. See Disclaimer and Impressum.