1 of 26

Eclipse SDV AI SIG – Proposal Detail

SIG Proposal Backup – John Stenlake (Microsoft)

1

2 of 26

Context (can be skipped)

2

3 of 26

Generative AI�is changing the automotive industry

3

4 of 26

Generative AI�

Generative AI engine

Memory & Context

Universal interface

Ask me anything…

4

5 of 26

Agents are an extension of GenAI models, giving it the ability to interact with its surroundings and providing additional context for its role

A2A

Tools Interface

Agent

Tool Control

Tool Control

Profile, Goals and Instructions

Model based Reasoning & Planning

Local Memory

User

Main building blocks

Memory interaction Context

Model based Reasoning & Planning

Profiles, Specific Know How

Utilizing industry standards

Agent

MCP

A2A

User interaction

MCP

Agent Interface

Agent

Agent

A2A

Flexible Data structures

5

6 of 26

Generative AI supports every stage of the V-Model

data architecture

Use Cases

Agents can collaborate on tasks

Agents support within existing development tools

Agents support across tool boundaries

Service

Production

Configuration Vehicle / Product Structure

Regulations & Standards

Homologation

Mechanical

E/E

Software

Requirementsmgmt.

Architecture

Integration & Testing

Ideation

Build on emerging Standards

6

7 of 26

Multi Agents can collaborate on complex Tasks

Single Agent

Agent

Tools

User

Single Tasks / Department

Agent

Agent

Tools

Agent

Tools

Agent

Tools

User

Agent

User

Multiple Tasks / Departments Interaction

MCP

A2A

User interaction

7

8 of 26

Single Focus Area Use Cases

data architecture

Activities

Coding

Requirement Level

Modeling

Service

Production

Configuration Vehicle / Product Structure

Regulations & Standards

Homologation

Mechanical

E/E

Software

Requirementsmgmt.

Architecture

Integration & Testing

Ideation

Issue Tracking

8

9 of 26

Agent Collaboration across multiple Focus Areas

data architecture

Main building blocks

Integration Agent

Ticket Agent

Architecture Agent

Service

Production

Configuration Vehicle / Product Structure

Regulations & Standards

Homologation

Mechanical

E/E

Software

Requirementsmgmt.

Architecture

Integration & Testing

Ideation

Triage Agent

9

10 of 26

AI Assistant support in Requirement Management

AI Agent use case

Requirement Management

System Level

Requirement Level

Extraction and Processing

Quality & Compliance Improvement

Classification and Organization

Traceability Management

Gap Analysis

1

2

3

4

5

Stakeholder Communication

Reuse and Pattern Recognition

7

8

Change Impact Assessment

6

Extraction and Processing

  • Automated extraction of requirement-like statements from unstructured documents
  • Natural language processing to transform stakeholder language into formal requirements

1

Quality & Compliance Improvement

  • Identification of vague terms
  • Suggestions for improving testability
  • Checking compliance
  • Suggesting corrections to meet quality guidelines

2

Gap Analysis

  • Detection of conflicting requirements
  • Analysis of requirement sets for completeness
  • Suggestion of additional requirements to address identified gaps

5

Stakeholder Communication

  • Generation of tailored requirement summaries for different audiences
  • Creation of visualizations to communicate requirement relationships
  • Drafting of clear communications about requirement changes

7

10

11 of 26

Sub Process - Ticket Triage

Triage Agent

Initial Data Analysis & Hypothesis Forming

  • Orchestrate Requests
  • Identify relevant tickets
  • Review related Architecture elements
  • Understand current Integration state

  • Formulate initial Hypothesis

Process Flow

Symptom Analysis, match historical data, recommend additional data collection

1

Architecture Agent, provide system topology and communication path

2

Data Analysis filter relevant data from noise

4

Integration Agent sub system changes & known issues

3

Hypothesis formation ranked cause analysis, recommend responsible team

5

Architecture agent

System Architecture Analysis

  • Identify affected subsystems
  • Map symptoms to vehicle architecture
  • Trace communication paths

2

5

Integration Agent

System Integration

3

  • Provide subsystem change context
  • Provide known issues
  • Correlate supplier / subsystem

Ticket Agent

Symptom Collection & Analysis

  • Document error manifestations
  • Determine conditions
  • Assess reproducibility
  • Analyze issue patterns

1

4

11

12 of 26

Eclipse SDV AI SIG

12

13 of 26

Proposal: Rationale and Objectives for Forming the AI SIG

Unifying AI Experts

The AI SIG would bring together experts and enthusiasts to foster collaboration and innovation in AI technologies.

Knowledge Sharing

The group would promote sharing of insights and research to advance understanding and application of AI to SDV

Developing Best Practices

AI SIG focuses on creating frameworks, standards and best practices to address AI deployment challenges effectively.

13

14 of 26

Why Now

  • Rapid GenAI adoption pressure in automotive tooling
  • Fragmented agent / tool interoperability approaches
  • Opportunity to shape neutral, open reference patterns
  • Based on industry standards (MCP, A2A)
  • Ecosystem readiness ahead of 2026 investment cycles

14

15 of 26

Why a SIG?�

Need a Focus point for AI-related topics

While individual projects might also look at aspects of AI a joined-up approach can help accelerate AI benefits across the whole Eclipse SDV domain

Broad Scope

There are multiple aspects to AI relating to SDV to be considered – AI for Toolchains, AI in-vehicle stacks, Autonomous functionality, distributed Agents – which go beyond the scope of individual projects.�Initial focus is on Toolchains.

Flexible Governance

A SIG allows us to meet with individuals who are outside Eclipse membership (although they cannot vote or direct meetings). This enables close cooperation with other communities as appropriate. A cooperative approach with COVESA is proposed due to common interest and the ability to build broader support.

15

16 of 26

Vision Statement

Enable interoperable, agent-augmented automotive development tooling across the V-Model through open, practical service capability specifications and reference implementations.

16

17 of 26

Proposed Initial Focus: AI Enabled SDV Toolchains�

AI Toolchain Design Vision

Focus on designing AI toolchains that integrate seamlessly to support complex workflows efficiently. Specifically:��Enable interoperable, agent-augmented automotive development tooling across the V-Model through open, practical service capability specifications and reference implementations.

Implementation Strategies

Explore practical approaches for implementing AI toolchains to maximize performance and scalability. Seek collaborative opportunities to substantiate and demonstrate these approaches including:�� - collaboration with COVESA on the above goal

- collaboration with Digital.auto and their SDV lab

- potential collaboration with EU funded projects

Impact on AI Adoption

Examine how AI toolchains accelerate SDV and AI adoption across diverse industries, driving innovation and efficiency.

17

18 of 26

Foundational Principles

  1. Open and implementation-neutral
  2. Incremental & reference-driven (specify and demonstrate)
  3. Collaborate with other industry consortia (COVESA), and also other projects (digital.auto, EU funded initiatives) where beneficial
  4. Align with existing standards (AUTOSAR, ISO 26262, ASPICE) where beneficial
  5. Minimize complexity; focus on pragmatic adoption
  6. Vendor + OEM co-shaping

18

19 of 26

V-Model Focus areas (simplified)

Stage

Why It Matters

Agent / AI Leverage

Requirements

Alignment & traceability

Summaries, gap suggestions

Architecture

Consistency checks

Model ↔ requirement validation

Implementation

Artifact generation

Code/model scaffolds, impact hints

Integration & Validation

Test & assembly

Test gap clustering, planning

Homologation

Compliance evidence

Automated collation

main focus areas

19

20 of 26

Initial Focus Area (Pilot)

Pilot Scope:

Establish the working pattern

Three pilot phases (repeated for each V-Model stage):

  1. Curate & prioritize key requirement scenarios
  2. Specify neutral MCP/A2A service capabilities
  3. Build and validate a minimal MCP server + sample client�

Then: review, revise, repeat

Three-Phase Pilot Pattern

20

21 of 26

Sample Timeline (Illustrative)

Month

Milestone

Nov 2025

Kickoff + finalize pilot scope (Requirements Management??)

Dec 2025

Use case documentation & consolidation

Jan 2025

Draft service (pilot) capability spec v0.1

Feb 2026

Reference implementation alpha

Mar 2026

Vendor feedback + spec v0.2

Apr 2026

Broader outreach + 2nd stage selection

21

22 of 26

Deliverables (Foundation)

The initial phase deliverables aim to establish a solid foundation for the working group:

  1. Tool landscape analysis (initial)
  2. Requirements management use case pack
  3. Working group operating procedures

Ongoing the working group will maintain and grow a set of living outputs:

  • Updated tool category mapping (quarterly)
  • Use case library expansion
  • Best practices & integration guidance

22

23 of 26

Success Indicators

  • Technical: Reference MCP service running for pilot category
  • Adoption: At least one vendor trial / feedback cycle
  • Ecosystem: Growing participation across roles
  • Quality: Clear specs + example-driven documentation

23

24 of 26

Request for Steering Committee Approval�

Community Support and Interest

We’ve received statements of support from: Bosch, ETAS, AVL, Elektrobit, Kinnovia, T-Systems, Harman, UL Solutions, TTTech, useblocks, and Microsoft.

In addition, the COVESA parallel group has support from Rivian VW, VXLabs, Ropix, SDVerse, Modernize.

Kick off SIG Activity

Subject to Steering Committee approval, we propose to initiate workshops as soon as possible to agree scope and action plans and initialize SIG activity.

Steering Committee Request

I’d like to thank the committee for their attention and consideration, and would like to formally request approval of the proposed SIG.

24

25 of 26

Plans for Participation, Collaboration, and Next Steps

Community Participation and Interest

Encourage active community involvement to build engagement and shared ownership in the AI SIG.

Finalize Proposal

Outline clear goals and deliverables and finalize written proposal for distribution in mailing list, and propose to the Technical Advisory and Steering Committees

Kick off SIG Activity

Subject to support, initiate workshops to foster knowledge exchange and initialize SIG activity.

Interested? Please get in touch!

25

26 of 26

Thank You

Questions, feedback, offers to contribute welcome.

Contact: john.Stenlake@microsoft.com

%

26