Accelerating Software-Defined Vehicles: SDV-EG Updates & New AGL SDV Referenc... Jerry Jiancong Zhao

### Overview In this presentation, Jerry Jiancong Zhao from Panasonic Automotive Systems, leader of the Automotive Grade Linux (AGL) Software-Defined Vehicle Expert Group (SDV-EG), reviews progress made throughout 2025 and outlines upcoming directions for 2026. The session illustrates how software-defined principles have expanded from consumer technology to automotive platforms, driven by the need to decouple software from hardware and establish hardware-agnostic virtualization and OS-agnostic application frameworks.

### Key Topics - **Evolution of SDV and Architecture**: Zhao contextualizes SDVs within the broader "software defined everything" (SDX) paradigm, where continuous software updates, subscription capabilities, and over-the-air (OTA) updates replace fixed hardware functions. Automotive architectures oscillate between centralized and distributed models, emphasizing decoupling applications from operating systems and decoupling operating systems from underlying hardware. - **Device Virtualization via VirtIO**: The SDV-EG prioritizes VirtIO across four distinct implementation models: classic hypervisor-based virtualization, non-hypervisor VirtIO deployments (leveraging containerized setups and priority/resource management methods developed with the Instrument Cluster Expert Group), cross-hardware/multi-ECU virtualization, and cloud virtual environments. - **Unified HMI 2.0**: The group extended VirtIO-GPU to function over networks via Remote VirtIO-GPU (RGPU) alongside a distributed display framework to coordinate application rendering across multi-display setups. Version 2.0 refactors Remote VirtIO-GPU to minimize windowing/OS dependencies across Linux, AGL, and Android, and integrates gRPC protocols for multi-language application support and life-cycle management. - **Cloud-to-Vehicle and Data Capabilities**: The SDV-EG is exploring cloud-to-edge parity using virtual targets (such as Virtual SC/SKIP) to execute AGL or Android workloads on cloud ARM servers (e.g., AWS Graviton) with identical VirtIO guest interfaces without binary changes. Simulation back ends connect to ROS 2, model-based designs, or physical hardware. Collaborations with EMQ aim to enable closed-loop vehicle-to-cloud (V2C) data flows.

### Technical Details - **AGL SODA Initiative**: Serving as an incubator and integrator across AGL and Linux Foundation projects, the AGL SODA initiative adopts an architecture featuring Dom0 (incorporating Dom0-less concepts under discussion with the Xen community) and DomD (hosting VirtIO back ends, with Linux and Zephyr implementations by Tokida from Fujitsu under evaluation). Supported guest domains include Android Automotive (AAOS), AGL In-Vehicle Infotainment (IVI), gateways/graphical stacks, and real-time domains running Zephyr or Real-Time Linux for ADAS and instrument clusters. - **VirtIO CTS and Verification**: A new VirtIO Compliance Test Suite (CTS) approach is being introduced to verify back-end VirtIO implementations. - **Q&A Insights**: Zhao explained that binary compatibility in the cloud is achieved by pairing ARM cloud servers with standard VirtIO guest device interfaces. Additionally, while container orchestration (e.g., K3s) and OTA maintenance on edge nodes present networking and monitoring challenges, AGL is evaluating cloud-edge trade-offs. The SDV-EG is also tracking Linaro and Google's upstream efforts on VirtIO-message for heterogeneous OS communication.

### Takeaways The AGL SDV Expert Group is transitioning beyond conventional hypervisor paradigms to multi-ECU, containerized, and cloud-native automotive deployments. By integrating projects like Xen, Zephyr, Real-Time Linux, and gRPC, AGL aims to standardize open-source frameworks that decouple software lifecycles from proprietary automotive hardware.