AI and Machine Learning

#wearesubstorm | Jonas Bergqvist

He builds smart cars with the help of AI and machine learning.

Let us introduce Jonas Bergqvist – system developer by day and robot master builder by night. A man who is not afraid of using loads of components and different techniques to build what he wants.  We sat down with Jonas to talk about his latest construction – JetBot the smart car. JetBot drives itself, stays inside the white lines and stops when the traffic lights turns red.

I wanted to learn more about AI and machine learning in a playful but still very hands on way.
Jonas Bergqvist


Nickname: Jonbass
Age: 46
Lives: Skelleftehamn
Started working at Substorm: 2020
Languages: Swedish, english and C++
Favorite robot: Love the Boston Dynamics Atlas Robot, but most of the Robots from Boston Dynamics are awesome.
Bucketlist destination: French Polynesia

– It’s both visual and very concrete to work with a physical object because I get to see the response in real time. And it’s really cool to be able to do everything myself – from gathering all the data, process it and use it to make the car smarter. It has given me a deeper understanding of the relationship between hardware and software, that I already have been able to put to use in my work related projects.

But the road to a working prototype has not been without challenges. One of the biggest ones has been to secure the power supply for all the different components that drives the car without making the vehicle too heavy.

– I tried with AA batteries connected to powerbanks but that turned out to be way to clumsy. In the end, I ended up with 18650-batteries, like the ones used in flashlights. They are extremely powerful and with two of them, together with circuits, I can power the car without problems.


– I also had some issues with getting the remote desktop and VNC Viewer to function in a solid manner and run smoothly. A lot of the times it’s been very error heavy and slow to work that way – so a lot of the work has been done trough SSH and the console it self.

With cheap but great hardware like Raspberry P! and Ardunio, the price tag on this kind of project manages to stay around 100 dollars – with great results. The development process has been agile and contained a lot of different stages – some that attracted the attention of even the family’s youngest member.

– My daughter helped me build the road and the environment around it by using figurines from Star Wars to simulate what happens in traffic, both at the traffic lights and as obstacles and pedestriants. I loved having the kids engaged in this project!

Timewise it’s been many long hours, especially when my initial car later turned in to three differens ones.

– I’ve built and picked apart, over and over again. And again. The picture recognition and the deep learning is what took the longest to get right since there’s a gazillion different combinations. JetBot took me about eight hours to get rolling – the other cars took a lot longer. In total I’ve spent way more that 50 hours getting all of these done.

Tech used for Autonomous Car I & II – Robot Car

  • 2x Chassi plates (Top & Bottom)
  • 4 DC Motors (5– 10V)
  • 4 Wheels
  • L298N Motodriver
  • Arduino Uno Rev3
  • Raspberry PI 4
  • Rapsberry PI Camera Module
  • Raspberry P Camera cable– 60cm
  • Breadboard
  • Extra cooling for Raspberry PI
  • 18650 Batteries
  • Powerbanks


  • Raspbian OS for Raspberry PI
  • Arduino IDE / Cloud IDE
  • Geany IDE (c++ / c)
  • OpenCV
  • Libjpeg
  • Wiringpi
  • Libavcodex
  • RaspiCAM
  • Remote Desktop
  • VNC Viewer