How removes one of the big obstacles in ADAS / AD development

How removes one of the big obstacles in ADAS / AD development

Jürgen Daunis

In the past few years, we were able to observe the rapid development of Advanced Driver Assistance Systems and Autonomous Driving solutions. Bringing a safe ADAS / AD system on the road, however, has often proven to be a challenging, expensive and complex undertaking., a dSpace company, is an expert in the field of ADAS / AD development delivering »data as a service« on a cloud based platform. We accelerate complex ADAS / AD programs by making the delivery of high-quality ground truth annotations scalable and commercially feasible through Artificial Intelligence and Machine Learning.

The levels of driving automation
Defining driving automation, SAE International has introduced the differentiation of 6 consecutive levels (0 – 5): Accordingly, up to level 2 we usually talk about Advanced Driving Assistance Systems (ADAS) and starting with level 3, we talk about Autonomous Driving (AD).

In level 0 – 2 of driving automation, the driving responsibility and liability mainly remains with the driver while above level 2, the manufacturer will enter the liability ownership discussion: In all levels, the manufacturer must prove and document that appropriate and sufficient testing and validation of the systems has taken place. With increasing complexity and liability risks in higher AD levels, the need for testing and validation grows as well.

Validating Autonomous Driving functions
To examine whether the sensors and perception models of autonomous systems correctly and reliably detect the vehicle’s environment under real-world conditions, validation projects are a necessary requirement to ensure everyone’s safety on the road. As part of this process, it is common to conduct drive tests that typically involve driving hundreds of thousands of kilometers and generating large volumes of annotated data on the vehicle’s surrounding. Executing these large scale validation projects naturally comes with a lot of quality, time and cost challenges.

The importance of ground truth data
For one, in order to run a validation project successfully, you require high-quality validation data: In the development and validation phase, the devices under test (the parts that are build into the car) are often compared with a separate independent system (e.g. a rooftop box) to ensure it is working correctly. The recording of this reference system will then be used to generate the »ground truth data«.

By comparing the device under test against ground truth data, you can test its performance and accuracy. Furthermore, you can validate how well the assisted and autonomous driving functions recognize the environment. Needless to say, having a diverse and representative dataset that covers standard scenarios as well as edge cases and critical situations is paramount. Therefore, being able to reliably deliver high-quality data that meets ground truth quality expectations, is extremely important if you want to find out if a perception model works properly under real world conditions.

However, staying within an affordable budget and time frame while generating these ground truth annotations has proven to be a significant industry challenge.

How removes one of the big obstacles in ADAS / AD development

Accelerating validation through Artificial Intelligence and Machine Learning
As a result of applying Artificial Intelligence and Machine Learning, we are able to overcome this challenge by decreasing the manual annotation efforts. In consequence, we can deliver more ground truth annotations with better quality in a shorter time. This allows us to accelerate the validation projects of our customers significantly and, accordingly, be more cost-effective.

The importance of domain competence, data quality control and business decisions
Operating solely through a »generic platform« perspective, however, is often not sufficient to make these validation projects successful. In order to handle the complexity of refining the automation to the right level, extensive domain competence is essential: The specific sensor setup, dealing with sensor calibration issues, the Operational Design Domain (ODD) and specific functional requirements – applying Artificial and Machine Learning technology correctly and efficiently requires thorough expert knowledge of the domain we are operating in.

On top of that, we have recognized the importance of implementing systematic quality control and methodology to ensure consistent data quality. In addition, we ensure to make balanced business decisions when determining where to apply automation based on commercial feasibility.

Scaling and industrializing annotation production systems
Another indispensable requirement is the scalability and the industrialization of the production systems for large data volumes: The ability to automate large quantities of data is tightly linked to the infrastructure the services are running on. Hence, strong knowledge of cloud infrastructure and flexible scaling capabilities are necessary to industrialize the annotation production while also being efficient.

Through our platform, we are combining IT, Cloud as well as Artificial Intelligence and Machine Learning automation technology along with extensive services, processes and domain competence. This has resulted in fast delivery times and the reliable generation of high-quality ground truth data. Moreover, it allows for feasible project costs and the necessary project scalability.

Offering an end-to-end data pipeline
In addition, delivers complete end-to-end data pipelines for our customers to address all stages of the development process for Autonomous Driving. Accordingly, we collaborate with our mother company dSpace on a data-driven software development approach but also offer an independent, end-to-end annotation pipeline to our customers. As a result, we are able to make the data pipeline for the generation of high quality training and validation data easier, faster and more efficient.

How removes one of the big obstacles in ADAS / AD development

Making ADAS / AD projects feasible at a large scale
To conclude, large-scale validation of Autonomous Driving functions is essential to ensure safety on the road and compliance with liability requirements for ADAS / AD manufacturers. In order to be able to process these high amounts of data, applying Artificial Intelligence and Machine Learning is indispensable: Decreasing manual annotation efforts allows for more high quality ground truth annotations to be delivered in a shorter time and, consequently for larger data volumes to be annotated.

A high level of domain competence, combined with the application of commercially viable business decisions and methodology, is crucial for refining the automation. Moreover, flexible end-to-end data pipelines can simplify and accelerate the implementation and homologation of Autonomous Driving functions significantly.

We, at, are experts when it comes to generating ground truth data in the field of ADAS / AD development and large scale validation. Our expertise results in faster data delivery and a significant cost advantage for our customers without having to compromise on reliable quality – making ADAS / AD projects scalable and commercially feasible as a result.

Dr. Wolfgang Eckelt, High Performance | Top Company Guide