Original article was published by Jeremy Cohen on Artificial Intelligence on Medium
Choosing a ROS Platform—
The Raspberry Pi has always found a community of edge Engineers who wish to build Do It Yourself projects.
What is it? It’s a mini-computer that is small, performant, and relatively cheap.
As I am building a new ROS course, I got interested in the Raspberry — Can this small device run ROS, embedded Computer Vision software, and be used as a remote platform for my students?
I looked into the Raspberry Pi 4 for a few weeks, I tried to understand how to use this tool the best I could… and how it would be a great use for my problem.
I also tried to see if other more appropriated devices such as the Nvidia Jetson Nano could be better suited. Here’s what I found —
Raspberry Pi 4 vs. the rest
If we are to build a Do It Yourself project, or something like Edge Computer Vision, a lot of possibilities appear.
- For example, I recently wrote about VPUs and the Intel Movidius NCS. That’s a super powerful USB Key that can be plugged to a computer and provide a lot of processing power.
- More recently, the Nvidia Jetson Nano has been released and got a lot of attention.
- Soon, the OpenCV AI Kit will be there as a revolutionizing tool.
The last two computers have their own SDK; which means they won’t work right away.
Most Computer Vision and edge devices are very good when it comes to running an OpenCV program, but are they good when we have to work with Robotic OS, LiDAR point clouds, and GPIO sensors? Do they provide the possibility to plug any sensor and just make it work?
This is one of the reasons I looked into the Raspberry Pi. It’s running Linux, has tons of GPIO possibilities, and can be used immediately! It’s more of a tiny computer than the rest because it wasn’t built for AI.
It also has a lot of processing power, and the new 8Gb RAM version makes it super powerful! That, and the low price, are key elements.
Imagine students enrolled in my course want to buy it and try it themselves, a Jetson Nano would be a 200$ investment. The Raspberry Pi 4 costs about 100$, which was great for my use.
If you have the budget, you can probably go for a Jetson Nano or even better, but in our case, the Raspberry Pi 4 can totally work.