In the rapidly evolving world of Software-Defined Vehicles (SDV), the leap from development to deployment is often hindered by the sheer complexity of safety validation. At Vayavya Labs, we specialize in “silicon-to-system” engineering, focusing on digital twins, functional safety, and embedded full-stack development. We are excited to present our SOAFEE blueprint, a practical approach for a cloud-to-edge testbench designed to validate Autonomous Driving (AD) stacks at scale.
The Challenge: Beyond Real-World Testing
Currently, ADAS and AV development relies heavily on real-world testing, which is both expensive and time-intensive. Physical road tests cannot feasibly cover the near-infinite variety of edge-case driving scenarios required for safety certification.
To bridge this gap, the industry needs a scalable, cloud-native virtual validation pipeline. The goal is to achieve behavioral parity—ensuring that the software performs exactly the same way in the cloud during testing as it will in the actual vehicle.
The SOAFEE Architecture: A Unified Framework
Our blueprint leverages SOAFEE’s Edge Workload Abstraction and Orchestration Layer (EWAOL) to provide a containerized Linux foundation. This allows the same ADAS/AV workloads to be deployed seamlessly across cloud environments and embedded Arm hardware.
The architecture implements a distributed Software-in-the-Loop (SIL) testing approach using two distinct compute environments:
- The AD Stack (ARM64): Running on an AWS Graviton (G5g.4xlarge) instance, this layer hosts the Open AD Kit, which includes Autoware (for perception and planning) and a Vayavya-developed Stack Server to manage communication.
- The Simulation Environment (x86_64): This side utilizes NVIDIA GPU acceleration for high-fidelity rendering. It features the Carla simulator for vehicle dynamics and the Esmini engine for scenario orchestration.
Seamless Communication and Middleware
The core of this blueprint is the Vayavya-developed middleware, which utilizes gRPC (Google Remote Procedure Call) for bidirectional data exchange. This enables lock-step execution, ensuring that the simulation and the AD stack remain perfectly synchronized.
Key data exchanged includes:
- Sensor Data: High-frequency LiDAR, camera, and radar streams.
- Actuation Commands: Steering, braking, and acceleration instructions from the stack.
- Ground Truth: Reference data used to evaluate algorithm performance in real-time.
Putting Theory into Practice: The ALKS Demo
To demonstrate the effectiveness of this blueprint, we conducted a simulation of an ALKS (Automated Lane Keeping System) cut-in scenario.
- Configuration: The user selects a test setup, such as Autoware running on Graviton and Carla on x86_64.
- Sensors: A suite of virtual sensors—including LiDAR, GNSS, and IMU—is configured on the “Ego” vehicle.
- Execution: Using the ASAM OpenScenario standard, the system triggers a scenario where another car abruptly enters the lane.
- Result: The vehicle successfully identifies the threat and handles the cut-in autonomously, allowing developers to capture results for further debugging and analysis.
Conclusion: A Shared Future for V&V
By closing the validation gap, this blueprint provides stack developers with the confidence to deploy software safely and reliably on real vehicles. As an active member of the SOAFEE community, Vayavya Labs is proud to contribute to the standardization of verification and validation for Automated Driving Systems.
This solution enhances the confidence of stack developers in deploying their software safely and reliably on real vehicles. Through this initiative, we aim to address critical challenges in the verification and validation of Automated Driving Systems (ADS). We are honoured to contribute as an active member of the SOAFEE community, collaborating closely with various working groups, and are proud to announce the release of our first SOAFEE blueprint.
Contact Vayavya Labs at sales@vayavyalabs.com to understand how you can leverage this blueprint.
