STRADVISION and aiMotive Combine Perception-Driven Scenario Understanding and Scalable Simulation for ADAS Validation
STRADVISION and aiMotive have announced the results of a joint proof-of-concept that illustrates an advanced workflow for developing Advanced Driver Assistance Systems (ADAS). This collaboration demonstrates how real-world vehicle fleet data can be converted into high-fidelity synthetic environments, which are indistinguishable from actual sensor data, thus validating SVNet perception outputs in real time on cloud infrastructure.
The partnership addresses a significant challenge in scalable ADAS development: transforming real-world fleet recordings into hyper-realistic, simulation-ready synthetic environments. STRADVISION provided its SVNet perception platform, which has deployed over five million production units globally. This platform allows for an Operational Design Domain (ODD)-aware interpretation and scenario extraction from Korean road recordings, focusing on perception-critical driving scenarios to enhance validation workflows. aiMotive's World Extractor then takes these perception-derived scenarios and raw data, applying neural reconstruction to develop detailed 3D environments using Gaussian Splatting. The process generates synthetic sensor data that closely resembles the original footage.
The synthetic datasets were created using aiSim, recognized as the world's first ISO 26262 ASIL-D-certified automotive simulator. AiFab is capable of producing diverse scenario variations at scale, effectively covering complex edge cases that might be difficult to capture through conventional real-world data collection. The system can incorporate a variety of 3D assets, including dynamic actors like vehicles and pedestrians absent in the original recording, as well as static elements such as traffic signs and road infrastructure, allowing for the creation of numerous unique scenes. The entire workflow, from ingesting raw sensor data through neural reconstruction to scenario generation and synthetic data export, was validated at scale on cloud infrastructure.
This integration fosters a feedback loop between real-world perception and simulation, enhancing scenario coverage and facilitating more efficient deployment of ADAS and autonomous driving systems. The established feedback loop aims to minimize disparities between field testing and simulation-based validation processes, eliminating the need for manual 3D environment creation.
Insu Kim, Head of STRADVISION's Data Innovation Center, commented on the integration, stating, "Real-world driving data alone is no longer sufficient to scale validation for next-generation ADAS systems. Through this collaboration, we demonstrated how perception-driven understanding of complex road scenarios can be transformed into scalable simulation workflows, helping close the gap between field operation and virtual validation."
Szabolcs Jánky, SVP of Product Strategy at aiMotive, emphasized the importance of extensive virtual validation for automated driving: "We, at aiMotive, strongly believe that safe automated driving requires extensive virtual validation. This project provides proof of how two like-minded and agile companies can build and deploy an efficient, high-quality neural simulation pipeline for an end-to-end automated driving software."