14 March, 2023 Webmaster

The Software Defined Vehicle Ecosystem

In Track A Electronics & Architecture on May 24th at VECS 2023 we will have the pleasure of listening to Erik Coelingh, VP Product, Zenseact and Adjunct Professor Mechatronics at Chalmers University. Erik joined the Volvo Car Corporation more than 20 years ago, where he was responsible for Volvo’s first application of Automatic Emergency Braking and led the advanced engineering activities for Pedestrian Detection and Intersection Collision Avoidance. His work focuses on algorithm and system design for active safety features for collision avoidance by means of warning and automatic emergency braking and steering. He also works on the design of driver support features to take over part of the driving task, utilizing adaptive cruise control, automatic parking and autonomous vehicles. His ultimate vision is to design cars that can be driven fully automatically without incident. We got the opportunity to ask Erik a few questions before the event and he was kind enough to share his insights as well as giving us a teaser of his presentation.

 

What will you speak about at VECS 2023?

At VECS 2023 I will describe how software-defined vehicles allow for faster traffic safety improvements. By probing data from consumer vehicles, we can learn how people use active safety features, and how and where these features impact traffic safety. With this understanding we can adapt the software and update the vehicles in order to speed up the trend towards zero collisions.

How can we increase speed in software development?

The speed of software development is increased through automation. We are building and testing software continuously such that we can hit the road with a new software release every single day. At Zenseact we have taken all software development for active safety systems in-house, from computer vision to vehicle control, in order to have these fast iteration loops.

What are the latest innovations in this area?

Deep learning has revolutionized the perception capability of the vehicle. We have built an infrastructure that allows us to deploy neural nets at an industrial scale, e.g. through our GPU farm, data pipelines and data centers. We have shown that we now can quickly improve e.g. computer vision algorithms, by collecting and annotating relevant data. Based on this we foresee a huge performance improvement when we apply this to data collected from consumer vehicles around the world.

What new architecture tradeoffs are in focus?

There is the obvious trend towards more centralized compute where significantly more memory and computing power is available, as compared to older vehicles. In addition, redundancy of compute and actuation are important when moving to unsupervised technology.

Looking at the mobility race, do you think the tech giants and new entrants will dominate the scene, or can the established OEMs and their suppliers take advantage and create a new position in the eco system?

The race is still open. Winners will be the players that can quickly make the move toward software defined vehicles. It is about being able to build active safety and self-driving vehicle tech, being able to deploy it at scale and being able to update software regularly to grow customer value.