At the origin of this platform, SEGULA Technologies wanted to test its ability to integrate a complete set of key technologies into an autonomous vehicle and aimed to enhance the skills of its employees on the subject. The company was also able to organise an ecosystem of startups, capable of developing or providing these technologies.

A platform to learn how to navigate autonomously

The platform was designed like a valet parking service, except that the valet is not a person but an application developed by SEGULA Technologies!

‘The owner calls their vehicle parked in the car park using their smartphone, where they’ve installed the app, to request it to come and meet them,’ explains Julien Fouth. ‘The vehicle receives the call, “understands” it, and starts moving through the car park.

Manoeuvring in an enclosed space is not that simple, especially since there has been no prior mapping or GPS. ‘Artificial intelligence gives the vehicle self-learning capabilities, using onboard sensors, among other things,’ he continues, ‘enabling the autonomous car to navigate in a crowded environment, reach the toll barrier, exit the parking lot, meet the user who called it, and take them to their destination, while assessing the traffic situation depending on the weather, road conditions, etc. These functions have been tested on SEGULA Technologies’ closed circuit, with various routes and different vehicle call scenarios.

Modules operating through artificial intelligence

Since then, around thirty research and innovation projects have been carried out on the platform, focusing on issues such as location sensor processing, human-machine interface, and navigation. ‘We have worked with a car manufacturer to validate each function of an electric vehicle controlled by artificial intelligence,’ says Julien Fouth. ‘One of the experiments involves, for example, crossing a roundabout using AI modules.’

For this project, the chosen technology is based on “explainable artificial intelligence” methods and “reinforcement learning.” The reinforcement learning approach relies on deep neural networks.

‘Without going into too much detail, it involves learning the decisions to be made from experiences,’ he continues. ‘Each time the autonomous vehicle makes a correct decision, the environment provides a reward. Each time it makes a wrong one, there’s a penalty. This way, the vehicle optimises its decision-making behaviour.’ Around the roundabout, specific zones have been defined: observation zone, decision zone, danger zone…

A range of expertise

SEGULA Technologies mobilised, for this purpose, a range of expertise, including an AI designer who creates the model, defines the necessary dataset, and analyses the results; an integrator for electronic solutions on the autonomous vehicle; a data engineer to label the data (i.e., assign information to the data so that machine learning algorithms can better understand their meaning); a human-machine interface developer, in other words, dashboards that analyse information in real time…

“We have all these profiles at SEGULA Technologies, which allowed us to successfully carry out this project for our client,” says Julien Fouth proudly. “I had the opportunity to lead this project for two years, which enabled me to witness the exceptional power of artificial intelligence in decisionmaking by an autonomous vehicle. It remains a particularly innovative experiment to this day!”