AVL ZalaZONE recently opened their offices at the ZalaZONE Automotive Proving Ground in Zalaegerszeg, and as part of the inauguration ceremony on the 16th of November the guests had the opportunity to gain insight into some research projects and results that wouldn't be possible without the help of the research and innovation engineering team and  the AVL ZalaZONE crew. As part of the festivities we also presented a live demo: a realistic use case for the Central System software developed by BME Automated Drive Lab.

This software system is a high bandwith, low latency data streaming platform designed with a special focus on autonomous driving technology to enable communication between mobile and stationary agents of traffic systems.

We've already shown this system in action during the IV. Trilateral Conference in July, where we used it to set an SPT (Soft Pedestrian Target) on a collision course with a car to test the latter's internal emergency braking systems, and we were eager to show the improvements we have made to the system since then.

‍

This new demonstration on the 16th of November involved multiple parties:

• A single VUT (Vehicle Under Test) carrying the testing crew's full complement of steering robotics, connected wirelessly to the Central System, broadcasting its location, and listening for external triggers.
• A networked camera attached above the VUT's windshield, looking straight ahead, with a laptop computer in a separate location using its video stream to detect pedestrians in front of the vehicle, sending results to the Central System.
• An SPT (Soft Pedestrian Target) also connected to the Central System and listening for triggers.
• A desktop computer we have named Control, running MatLab Simulink, connected to the Central System, and orchestrating everything.

‍

The setup might be familiar to the returning reader, but instead of using multiple stationary detectors, we only had a single vehicle-mounted camera.

The shown scenario goes as follows:

1. The VUT launches a pre-planned trajectory, and speeds down its designated lane.
2. Control keeps track of the vehicle's location and speed using data it receives through the Central System.
3. Control uses this information to calculate how long it takes for the VUT to cross the SPT's planned path.
4. Control launches the SPT through the Central System when its expected Time to Collision matches the VUT's.
5. Control listens for messages from the laptop broadcasting locations of pedestrians in front of the VUT, detected using the camera image with an off-the-shelf machine learning model.
6. When the SPT is detected entering the collision zone, Control triggers the brakes on the VUT and stops it, preventing the collision.

Almost all communication took place through the Central System's interfaces, except the video connection between the car-mounted camera and the laptop computer, but there are already planned features that wouldn't just make sending video through the Central System possible, but preferable, further enabling any programmer to implement our libraries into their data processing pipelines quickly and painlessly.

The demonstration was challenged by the cold and rainy weather, but it worked flawlessly and no dummies were hurt.