Abstract This paper presents a novel, three-layered system architecture designed to enhance the reliability and safety of autonomous vehicles by dynamically adapting to contextual changes. The architecture consists of the context layer, reconfiguration layer, and architecture layer. The context layer extracts environmental and operational data, such as weather, traffic, and user preferences, and derives requirements from these inputs. The reconfiguration layer uses these requirements to... This paper presents a novel, three-layered system architecture designed to enhance the reliability and safety of autonomous vehicles by dynamically adapting to contextual changes. The architecture consists of the context layer, reconfiguration layer, and architecture layer. The context layer extracts environmental and operational data, such as weather, traffic, and user preferences, and derives requirements from these inputs. The reconfiguration layer uses these requirements to plan adaptive actions, such as selecting software applications, assigning redundancy levels, and optimizing hardware deployment. The architecture layer then implements these changes on the vehicle’s computing infrastructure, ensuring operational safety through monitoring and validation. Two illustrative use cases demonstrate how distinct driving scenarios—like premium highway rides or budget urban rides—influence application needs and redundancy levels. The application placement problem, which involves mapping applications to computing nodes while optimizing for constraints like CPU demand and safety, is tackled using techniques such as linear programming and reinforcement learning. System selfawareness is maintained through continuous monitoring, enabling swift reconfiguration in the event of failures. The proposed approach is unique in its application of full-stack context awareness to autonomous systems, drawing inspiration from strategies in aerospace fault management. Future work includes implementing a simulator to validate the architecture’s real-world feasibility. Read More Read Less

Abstract This paper introduces Fusion, a next-generation software platform designed to meet the complex safety, performance, and architectural demands of autonomous driving systems. Fusion uses a microkernel approach and blends service-oriented with component-based designs to ensure modularity, scalability, and robustness. It features an Object Specification Language (OSL) to define object behaviors and interactions, which are instantiated and configured through manifests, enabling clear and flexible system... This paper introduces Fusion, a next-generation software platform designed to meet the complex safety, performance, and architectural demands of autonomous driving systems. Fusion uses a microkernel approach and blends service-oriented with component-based designs to ensure modularity, scalability, and robustness. It features an Object Specification Language (OSL) to define object behaviors and interactions, which are instantiated and configured through manifests, enabling clear and flexible system modeling. The architecture supports concurrency via execution contexts and offers transparent communication across distributed systems using a pluggable transport mechanism. Key challenges addressed include managing system complexity, achieving high performance with low latency, ensuring safety and fault tolerance in Level 4 autonomy, and securing sensitive data and communication. Design-time verification, enabled by Fusion’s modeling capabilities, aids in early error detection and system optimization. Fusion also incorporates built-in services for health monitoring, logging, and scheduling, supporting a wide range of deployment environments from vehicles to simulation and cloud platforms. Currently used by Autonomous Intelligent Driving GmbH (AID), Fusion lays a foundational framework for developing safe, efficient, and adaptable autonomous driving applications. Future work aims to enhance the platform’s predictability, real-time capabilities, and resilience to support continued evolution in autonomous vehicle technology. Read More Read Less

Abstract This paper introduces an adaptable demonstrator platform designed to simulate distributed agent-based automotive systems, addressing the rising complexity of software-defined vehicles. As automotive electronic control units (ECUs) consolidate, there is a shift toward modular software design with run-time adaptability, especially to meet fail-operational and safety requirements in autonomous driving. The proposed platform leverages a distributed, real-time simulation using the SimPy framework, enabling simulation of... This paper introduces an adaptable demonstrator platform designed to simulate distributed agent-based automotive systems, addressing the rising complexity of software-defined vehicles. As automotive electronic control units (ECUs) consolidate, there is a shift toward modular software design with run-time adaptability, especially to meet fail-operational and safety requirements in autonomous driving. The proposed platform leverages a distributed, real-time simulation using the SimPy framework, enabling simulation of task execution and network communication among ECUs via the SOME/IP middleware. A key feature of the platform is its agent-based graceful degradation approach, which reallocates resources from non-critical to safety-critical applications during failures. The demonstrator setup, implemented on Raspberry Pi devices interconnected via Ethernet, effectively models real-world failover scenarios. The middleware supports decentralized service discovery and publish/subscribe communication, enabling dynamic task reallocation and reconfiguration. Experimental results show strong alignment between simulated and actual network traffic, validating the platform’s realism. Its hardware-agnostic, scalable architecture allows simulation on various configurations and network topologies. While real-time constraints may limit scalability on low-power hardware, the platform provides a valuable tool for evaluating emerging distributed automotive strategies in both academic and industrial contexts. The study concludes that this flexible simulation environment facilitates rapid development and testing of robust, fail-operational automotive systems. Read More Read Less

Abstract The UNICARagil project aims to develop a cloud-based infrastructure supporting four types of fully automated vehicles—autoELF, autoTAXI, autoCARGO, and autoSHUTTLE—each with distinct functional requirements. Given the project’s scale, involving over 100 developers from 15 academic chairs and 6 industrial partners, agile requirement engineering becomes crucial. This paper introduces a structured methodology combining agile and classical requirement engineering to accommodate evolving requirements across varied use... The UNICARagil project aims to develop a cloud-based infrastructure supporting four types of fully automated vehicles—autoELF, autoTAXI, autoCARGO, and autoSHUTTLE—each with distinct functional requirements. Given the project’s scale, involving over 100 developers from 15 academic chairs and 6 industrial partners, agile requirement engineering becomes crucial. This paper introduces a structured methodology combining agile and classical requirement engineering to accommodate evolving requirements across varied use cases. A template-based survey was designed to gather 47 use cases, which were transformed into 42 system requirements using a pattern-based approach that emphasizes completeness, clarity, and traceability. These were then translated into system design components and cloud services, resulting in over 70 API endpoints.The design leveraged OpenAPI for endpoint specification, enabling automated code generation using tools like the OpenAPI Generator and bootprint to produce Flask server stubs and documentation. A GitLab CI pipeline ensures continuous integration by generating code and documentation upon each update. This iterative, modular, and scalable process allows rapid prototyping and adaptability, aligning well with the project’s dynamic development environment. The proposed methodology proves effective in managing complexity and supporting the agile development of a cloud system for highly automated and networked vehicles. Read More Read Less