Heiko Schilling, Stellantis
"Autonomous Driving Should Not Remain a Premium Feature"
Heiko Schilling is Senior Vice President of Software and AI Engineering at Stellantis.
Stellantis
Following BMW, Stellantis also wants to combine Level 2+ and Level 3 in one vehicle. Heiko Schilling, Senior Vice President Software and AI Engineering, explains how the ADAS system was developed and why it is not yet going into series production.
Mr Schilling, in preparation for this interview, it was noted that you shared a video on LinkedIn showing a Tesla, driving on autopilot, crashing head-on into another car. Can you guarantee that such an incident will never happen with the systems developed by Stellantis?
Tesla has Level 2 automation. The key is to ensure that customers clearly understand what the system can and cannot do. Level 3 means "Eyes off" and "Hands off", and it is crucial to communicate this difference precisely. Our AI learning algorithms are designed to consider complex scenarios and help prevent such incidents. At Stellantis, we strive to set the right expectations and provide technologies where safety is paramount.
The official statement says that STLA AutoDrive is an in-house development, while other Stellantis sources point to a collaboration with BMW. What is correct?
Stellantis developed the Level 3 system using an internal approach that reflects the company's commitment to innovation and advanced technology. The system also integrates valuable insights from collaboration with BMW and other partners, ensuring a seamless and innovative experience for our customers. We began development in November 2018 and have since developed the L2-Plus stack ('Eyes on, Hands off') and the L3 stack ('Eyes off, Hands off'). BMW launched the system in Germany last year. Our system is market-ready and can be introduced across Stellantis brands. We are currently analysing the business case for each brand.
Current models would be ready for series production
Are there still differences between the systems?
We started as a collaboration, but each OEM has made specific developments. Since BMW went into series production last year, we were able to further optimise our system. We use Mobileye's IQ6 cameras and are the first OEM to go to market with this technology. Additionally, we have made upgrades to the sensors as we had more development time.
Which current Stellantis models would be hardware suitable to enable Level 3 autonomous driving?
The configuration varies depending on the vehicle segment and cost. We are examining brand-specific models to see which can integrate the system. The standard STLA Brain architecture is available in various sizes and can accommodate the system. Existing architectures, particularly high-volume platforms, are also compatible. The previous version also runs on current and future architectures. The sensor configuration is flexible enough to work in both older and new Stellantis architectures. Our systems use a variety of advanced cameras and radars for Level 2 Plus. For Level 3, a Lidar system is additionally required.
How does STLA AutoDrive 1.0 differ technically and functionally from established systems like Mercedes-Benz's Drive Pilot or Ford's Blue Cruise?
The approaches are different. The topic of autodrive is very agile, and there are many developments in the field of AI models. Traditionally, AI models focused on the perception of the environment. Newer approaches take a holistic approach and also integrate motion planning, that is, translating the perceived environment into driving movements. Our system combines L2 Plus and Level 3 in one vehicle. This improves the user experience through a seamless transition between different automation levels. Tests show that drivers can currently use the system actively for about ten minutes during a daily commute of two hours. In the future, the system is expected to enable speeds of up to 95 km/h and be further optimised via OTA updates.
Autonomous driving for the mass market?
So the system is technically capable of enabling Level 3 up to 95 km/h?
Yes, we are already testing up to 95 km/h. Safety concepts are crucial: We need to demonstrate sufficient test kilometres and driving hours to ensure passenger safety. These safety tests and homologation in various countries are the current bottleneck. Technically, we could already introduce the system at higher speeds.
How quickly can Stellantis roll out the system to many models, including brands like Opel or Fiat?
Our goal is the democratisation of AutoDrive technologies. It should not remain a premium feature but should reach the mass market and our volume brands like Opel or Fiat through scaling.
What role does AI generally play in the development of autonomous driving systems?
AI models have existed for 20, 30 years. Such technologies initially develop slowly before reaching a disruptive phase. An example is Deepseek, which can operate similarly complex models with less computing power as OpenAI. Autodrive is driven by AI. The current focus is on perception, but it will increasingly also include motion control and other sensors. For example, interior cameras can determine whether the driver is attentive or has health issues. Such information must be integrated into AI models. We will see great progress in this area in the future.
What manpower is behind the development of the system and how large is the development team at Stellantis in general?
Our AutoDrive team consists of about 300 developers. In total, Stellantis employs around 4,000 developers worldwide in ten hubs, which span North and South America, EMEA, and Asia. In addition to AutoDrive, this also includes the development of the cockpit, the basic software, the operating system, the onboard electronics, as well as the backend systems and mobile applications.
About the person:
Heiko Schilling has been with Stellantis since August 2022 and leads the development of a software team to promote affordable, emission-free mobility. In his role, he is to drive the company's digital transformation and develop strategies for integrating modern software solutions into vehicle architecture.
Before his time at Stellantis, Schilling was, among other things, Vice President of Software Engineering and Head of Navigation at TomTom, where he was responsible for the development of navigation solutions for over 13 years. There, he led more than 50 software development teams worldwide and contributed to scaling the navigation business. Other positions in his career include Amazon, where he accelerated the company's global expansion as Head of the International Tech Expansion Team, as well as consulting and leadership roles at Greater Than and Gousto. He began his career as a scientist at the TU Berlin, where he developed innovative algorithms for navigation and traffic management systems.
This article was first published at automotiveit.eu
