The cloud-based system ZF Annotate leverages artificial intelligence to significantly expedite the validation of sensor data. It is claimed that this market-ready solution for passenger and commercial vehicles is 80% less expensive than manual systems, offering substantial cost reductions.
Modern automated driver assistance systems rely heavily on numerous sensors to accurately analyze a vehicle’s surroundings and execute safe driving actions. To enhance the development of these ADAS and AD technologies, ZF Group has introduced ZF Annotate, a cloud-based, AI-driven validation service.
On June 27, 2024, at the ZF Global Technology Day in Hanover, the group showcased this rapid, cost-effective, and market-ready service for both passenger and commercial vehicles for the first time.
Accurate and dependable data is crucial for developing advanced vehicle assistance systems. Various sensors like cameras, radar, lidar, and ultrasonic devices continuously gather data, helping to construct a three-dimensional view of the environment. These systems must promptly identify diverse objects such as vehicles, pedestrians, lanes, and traffic signs.
For the vehicle to operate on what is known in the industry as ‘ground truth’—the ultimate truth about its surroundings—the sensor data must be accurately processed digitally. ZF Annotate enhances this accuracy by comparing the vehicle’s sensor data with a high-precision reference sensor set known as the reference measurement. Operating independently from the sensors being tested and exposed to identical on-road information, ZF Annotate verifies this data.
Once collected, the data is uploaded to the cloud for analysis. Using AI, all relevant objects are precisely marked, classified, attributed, assigned unique IDs, and tracked if they move. This data forms a comprehensive environmental model—the ground truth. After this ‘annotation’, the software provides a highly precise comparative measurement, establishing ZF Annotate as an advanced AI-supported validation tool for testing and training modern ADAS/AD systems from Level 2+ to Level 5. Dr. Holger Klein, Chairman of the Board of Management stated that ZF annotate integrates the benefits of a robust, independent reference sensor set with a scalable cloud service that employs smart 2D and 3D tracking algorithms.
Traditional systems primarily used 2D annotation to validate reference data, mapping the environment in distance and horizontal angle. The 3D-capable ZF Annotate adds a vertical dimension to the data. Klein explained that this positions their solution as a unique, intelligent ‘cloud factory’ within the industry, streamlining a previously labour-intensive, expensive service in various ways. According to ZF, the precise and dependable data from ZF Annotate’s reference measurements can significantly hasten the development and refinement of complex ADAS and AD systems. Previously, system validation was laborious, time-consuming, and costly as reference data was manually annotated.
Artificial intelligence enables a tenfold acceleration of the validation process, cutting down annotation time from 12 months to just two, thereby saving customers both time and money, as highlighted by Klein.
Klaus Hofmockel, Head of Research and Development in the Driver Assistance Systems and Electronics division, added that ZF Annotate is capable of generating a ‘ground truth’ in the shortest possible time. The cloud-based service operates continuously, 24/7, swiftly completing the validation of reference data and doing so much faster than other market options without sacrificing quality.
Depending on customer needs, the reference sensors can be used directly on the test vehicle or in ‘Pursuit’ mode—a sensor set mounted on a separate reference data vehicle, useful for scenarios like parking evaluation without major adjustments to the tested vehicle. This adaptability makes ZF Annotate independent of specific sensor manufacturers and eliminates the need for costly modifications to customers’ high-tech test vehicles. Customers can incorporate this service into ongoing development projects, supporting the validation process most effectively and aiding developers in achieving flawless products.
Additionally, the reference data recorded is not limited to a frontal view. Depending on customer needs, the reference sensor set can provide a comprehensive 360-degree view, offering a detailed and precise depiction of the vehicle’s surroundings.
