Autonomous driving systems need comprehensive virtual images of the environment. These show the vehicle’s location and indicate any actions that are possible or necessary. Among other things, this requires a highly accurate localisation  function, which can be made even more precise by identifying landmarks using cameras and radar.

The result is that very large volumes of data have to be processed in real time at speeds of more than 20 Gbit/s. Over the course of a journey lasting a few hours, tens of terabytes of data will accumulate. This leads to the need for huge data storage facilities and high-performance computing systems of the kind that we have implemented in the form of an analytics platform in the Microsoft Azure cloud. We have also been working to overcome the challenges of data labelling.

Software and functions

• Trajectory planning 
• Lateral and longitudinal control
• Environment detection 
• Nvidia PX2
• Automotive Ethernet 
• Localisation
• Navigation Data Standard (NDS)
• Development processes based on ISO 26262 (functional safety)
• Model-based development
• AI-based development

Bertrandt has continuously improved the process of developing algorithms for environment detection sensors over the course of a number of internal and customer projects in the automotive industry. One key feature of these projects is the acquisition of expertise in artificial intelligence, machine learning and neural networks.

Another important aspect is the necessary in-depth knowledge of the camera, ultrasound, laser and lidar technologies used in automated vehicle functions. Bertrandt has been developing these technologies over several years for driver assistance systems such as adaptive cruise control and traffic jam, emergency braking and parking assistants.

On this basis Bertrandt has produced automated longitudinal control systems for production models which evaluate the speed limit, the shape of bends in the road and the distance from the vehicle ahead. The lateral control systems developed by Bertrandt include lane departure warning systems that are suitable for production use.

Technologies

• Domain architecture
• Automotive Ethernet
• Longitudinal and lateral control
• Systems engineering

Cars with automated and autonomous driving functions share data with back-end and cloud systems and with nearby vehicles. Bertrandt has completed a range of different internal and customer projects in the field of connectivity.

The number of sensors used for autonomous driving is continuing to increase, as is their resolution. As a result, the ability to manage large volumes of data for automation, digital test processes and integration is becoming increasingly important. Bertrandt also has expertise in developing and validating measurement systems, using reference sensors and running sensor evaluations.

In order to move closer towards the self-driving car, connectivity with other vehicles and with cloud and server structures is being enhanced. This allows up-to-date information about the weather and the traffic situation, for example, to be evaluated. Bertrandt is developing systems that enable the front-end, the back-end and the cloud platform to communicate with one another. These include both apps for mobile devices and web interfaces.

Developing cloud applications

• Microsoft Azure platform, Microsoft Data Science Virtual Machines, Microsoft Data Lake Store
• Data analytics
• Connectivity and back-end systems
• Device connectivity 
• TLS/PKI encryption

One important finding to emerge from our internal and customer projects is that a technology company such as Bertrandt, with expertise in all areas of vehicle development and design, has considerable advantages over specialist IT businesses.  More than just a knowledge of AI and IT is needed to develop algorithms for image recognition and sensor data analysis and to predict vehicle behaviour.

At Bertrandt, data scientists work in cross-disciplinary teams with specialists in vehicle design and dynamics in order to evaluate how a car will behave in a corner under lateral and longitudinal acceleration. Neural networks help to categorise objects on the road and adjust the vehicle control systems to accommodate scenarios such as steering, braking or taking evasive action. It is only possible to develop plausible and successful algorithms on the basis of a comprehensive understanding of the overall design and behaviour of the vehicle.

Internal and customer projects

• Innovation project to develop a level 4/level 5 car
• Park and Charge
• Lidar projects
• Driverless transport systems

As the level of automation in cars increases, so the methods used to test and validate new driving functions become more important. Bertrandt has been developing HiL, SiL and MiL tests on the basis of its extensive automotive expertise for 40 years. This allows us to combine the progress made towards developing autonomous vehicles with the testing of algorithms and the validation of new driving functions. We have developed user studies in driving simulators, simulation processes for virtual road testing, physical road trials and endurance and long-term tests. Bertrandt’s competence lies in the ability to test individual components, integrated systems and the interaction of all the systems needed for automated driving.

Validating autonomous driving functions

• Supplying test equipment (including a mobile solution)
• Test facilities for predictive pedestrian protection and emergency braking, for example
• Complying with the testing requirements of the 2018 EuroNCAP protocol
• Evaluating emergency lane keeping functions
   

Safeguarding by means of environment sensors

• Test benches for the safeguarding of radar and LiDAR sensors
• Sensor performance evaluation
• Bundling of expertise, testing know-how and a detailed evaluation
• Test kilometers saved due to the simulation of driving and environmental scenarios in the laboratory
• Acceleration of sensor development times

More about sensor technologies