Duckietown at BME — the debut

Source: Deep Learning on Medium


Go to the profile of SmartLab AI

So many ways out there to learn computer vision, control theory, robotics and machine learning. It can be hard to choose a way, a correct one, the one that answers all your questions, puts you through the hard parts easily, delivers results flawlessly and upgrades you from dummy to pro in a fraction of a semester. Well, Duckietwon certainly isn’t one of the easy ways, but we all know that with great efforts comes great knowledge, plus there’s lots of support on the foundation’s Slack channel and GitHub repositories. Also, it looks fun.

Source: Duckietown

More details on their website and in this video.

Preliminaries

Under the aegis of the Professional Intelligence for Automotive (PIA) project the Duckietown environment enforcing learning robotics and machine learning has arrived at our laboratory.

The PIA project is a collaboration project between Continental Hungaria Kft and Budapest University of Technology and Economics (BME) Faculty of Electrical Engineering and Information, Department of Control Engineering and Information Technology, and Department of Telecommunications and Media Informatics. The motivation of the cooperation project is to establish a strong relationship between industry and academics with the goal of delivering experiments and new approaches for automotive along with supporting student project and researches in BSc MSc and PhD programmes.

What’s in the box?

Our order contained the following:

  • 12 Duckiebots
  • 12 GPU Deep Learning Accelerator pen drives
  • 90 tiles for setting up multiple driving environments
  • 2 traffic lights
  • Many-many traffic signs
  • Lots of duckies
Our first Duckiebot

What are our plans?

We target to compete in the AI-DO 3 competition which will take place in December at NeurIPS 2019 in Vancouver, Canada. We already started in February to evaluate multiple reinforcement learning methods suitable for self-driving task and we narrowed down to 4 promising solutions which we now apply in the Duckietown simulator environment. More on this later.

The arrival

On Monday 08.04.2019 we received the package and the next day we assembled our first platform with a traffic light, some traffic signs and a Duckiebot of course. Here’s a time-lapse about the assembly which almost took us 6 hours.

And the field…

The first field

More writings will be posted during our quest until the AI-DO 3 competition. Stay tuned.