03.02.2021 - Final Presentation


Five challenging months after we started our company cooperation project, we are happy to share our final project outcome with you. Despite the wide range of problems we had to deal with, it has been an exciting endeavor for us, full of unexpected twists and turns. Though we were not able to implement every feature we initially thought of, we focused on the most important aspects and are now able to return a well-structured, easily extensible, and reliably working project back to PwC.

Before diving deeper into conclusions and reflections on individual project parts, this is an impression of our project, everything being put together:

https://youtu.be/z2g3LOJzuE4

You can see the red car being controlled by Abhik and the black car driving autonomously in "easy" driver mode. This poses no challenge for a professional like Abhik, and he finishes the five laps 0.15 seconds faster than the autonomously driving car. In the top right corner of our frontend you can see a standardized, top-down view of the current positions of the race cars. Three reference points on the race track enable a linear transformation of the relative position of the race cars in the video stream into an absolute position that shows the same perspective regardless of where the camera is put. The bottom of our frontend shows an all-time leaderboard saved in a Microsoft Azure Cosmos DB.

Putting everything together and visualizing all parts in one frontend with millisecond interval updates was one of the most challenging aspects of our project. Due to the limited capabilities of our frontend framework Dash, the final assembly of our code base required several workarounds which resulted in slight increases in latency. Surprisingly, the object detection of the race cars in the video stream was no computational bottleneck even when only run on CPU instead of GPU. This increases the portability of our project because no further CUDA configurations have to be made if the project is run on a different device.

In the end, this is how our project finally turned out in comparison to what we initially planned to do:

Conclusion

This project has been a great pleasure for us and we want to say thank you to PwC for handing us this project over for the time being and the Frankfurt School AI Lab team for the organizational support throughout the last months.

Yours,

Marius Andersen

Abhik Modi

Christopher Schmitz

Magnus Struckmann

30.11.2020 - Project Update #3


We are about to reach the final stage of our company cooperation project and are happy to share another project update with you.