Contracts for Services in Software-Defined Vehicles

Prof. Dr. Mattias Nyber, KTH Stcockholm / Scania

• Abstract

  The emergence of software-defined vehicles (SDVs) represents a paradigm shift in automotive software development — moving from system focus to service focus, where a service is not explicitly bound to an ECU or even to the vehicle itself. The result is open, dynamic ecosystems composed of distributed services. In this new context, service-based architectures and service-oriented systems engineering provide a foundation for integrating and evolving complex functionality across software, hardware, and even human interactions. A service view of all vehicle components — applications, actuator units, and sensor units — promotes a flat hierarchy where each element offers and consumes services. This enables horizontal service traceability in place of traditional vertical decompositions, fostering flexibility and modularity. However, such openness also demands rigorous specifications to ensure consistency and correctness across service interactions. This presentation explores how contract-based design and software contracts, in the form of pre- and post-conditions as used in ACSL (ANSI/ISO C Specification Language), can be employed to specify, reason about, and verify these service interactions. Formal contracts provide unambiguous interface definitions that capture both functional and non-functional expectations, enabling verification at multiple abstraction levels. Ambiguities that cannot be eliminated are instead managed explicitly through modeled dependencies.

Using AI Agents in Cotract-Based Design

Dr. Oscar Slotisch, Validas AG

• Abstract

  This talk introduces the concept of an AI (development) agent—an AI system used as a development tool rather than being embedded in the final product. In the context of ISO 26262, such agents are not part of the operational software but are employed during development, particularly in contract-based design.
  While standards for embedded AI are emerging (e.g., ISO 8800), AI agents used as tools can be assessed under existing ISO 26262 guidelines, specifically parts 8–11. This talk will define the Tool Confidence Level (TCL) from ISO 26262 8–11 and demonstrate how it applies to AI agents.
  A practical example will show how AI agents can be used to generate software contracts, including interfaces and assertions. Based on this, I will propose a set of requirements for using AI agents in a way that ensures compliance with ISO 26262, supporting safe and reliable development practices.

Additional speaker will be published soon

Short five-minute presentations to share research ideas