02 July 2023 18:11

Driverless - Three Team Perspectives (1/3 DE Aachen RWTH)

Category: Formula Student Driverless, Competition, 2023

For the 2022 magazine three teams gave us insights into the way they tackle the challenge of building an autonomous race car. With the magazine team working on full speed to finish this years magazine and FSG 2023 just around the corner we want to look back on this article ans post the full interviews the teams gave us. Starting with Ecurie Aix from the RWTH Aachen.

How do you perceive the development of the Driverless competition in the past years?

As members of the software group, we are of course pleased that Driverless is now firmly anchored with the competition and not just an additional option for the "big" teams in Formula Student - with all the advantages and disadvantages that entails.

It turns out that the victory is fought out among the teams that have the best understanding of the vehicle and work in a well-structured environment, not necessarily the best code in detail. So, we are happy that we are now working more with the whole team, who before, as probably in most other teams, often focused heavily, also for resource reasons, on the new EV car with their developments instead of maintaining the old car. 

As usual, Formula Student teams are quick to adapt, implement, copy, and develop: as soon as a concept emerges new to one team or industry, it is also quickly found in many teams in the following years. A decoupled suspension system is perhaps the best example here. But there has also been a clear trend in code about what methods teams are driving in the various blocks of the pipeline: YOLO for perception, GraphSLAM as a SLAM algorithm, Kalman filter for state-estimation, Delaunay triangulation for planning and model predictive control probably describes most software packages. The competition will therefore probably be decided at the ends of the pipeline in the future: Who knows his car best and has the fastest control on the one hand, and who can still provide sufficient data for his perception with the most minimal sensor setup. 

In this context, what is your opinion on lidar sensors: a supposedly easy entry into autonomous driving?

For new teams certainly the fastest and easiest way to get good data. But at the same time super expensive with a limited amount of suppliers. Especially now that there are so many teams, it is almost impossible to get a sponsorship for such a device, which is probably why not all teams can buy one. Also, the big teams often give the sensors to the new car. Perhaps one could have tied up here as organizers a package with a supplier, with central order, quantity discount, instead of as usual each team individually the mailboxes of certain industry enterprises to blow up... in principle however also a problem in some other departments. 

Personally, I find the trend towards lidar-only a shame, as working with cameras is very exciting, an active area of research and just more interesting for industry for cost reasons.

Especially for young teams, the ARWO is a perfect opportunity to learn something about autonomous driving - maybe also a role model for other departments?

We are very big fans of the workshop that Hamburg puts on every year - sharing new research areas, mistakes made in the past or disadvantages of certain concepts help both large and small teams. Everyone is free to choose what information they want to share with the others and get a little competitive advantage accordingly. I think such an event would also be interesting for various other disciplines, such as chassis or powertrain. The time frame, in a smaller, selected round allows for completely different conversations than perhaps at the Academy, which of course is always helpful. Perhaps it is possible to offer more workshops in the fall if there is enough willingness.

What do you wish for the future?

"Wireless Communication" with the vehicles, which would theoretically be allowed according to the current ruleset, is in our opinion a direction Formula Student should not go with autonomous driving. If you want to win, the probability that you will resort to it is given. But is this how you imagine autonomous driving? - No. A corresponding rule would at least increase the inhibition threshold not to do it, since control by the organizer is difficult anyway. 

I'm also very curious to see how the topic of "grip estimation" will develop in the coming years.

--
Interview Partner: Frederick Lockemann, Group Leader Driverless Software


Partners of FSG 2024