Building and Running the Pinball Autopilot on the Raspberry Pi

The source code lies and building happens inside the folder called pinballProject. For a fresh build, you need a not too old version of make, a C++ compiler (we only tested g++), an internet connection (for downloading the Kieler Compiler), the libraries wiringPi and opencv as well as the command-line tools mega-cmd and python2. The tool for mega.nz (and also python2) can be omitted (see below on how to compile without them), everything else except for opencv comes pre-installed with Raspbian. For opencv you need a 4.x version, 4.2.0 ought to work for example. Make sure to enable the option that the entries for pkg-config get configured. It can be built from source on the Raspberry Pi in a few hours, make sure not to use too many threads for compilation because of limited RAM, also perhaps increase the swap.

With all the pre-requisits met, you can use the Makefile to build everything.

For instructions on how to use the ./calibrate, see World Coordinate System.

Working without mega.nz (and python): On the pi, you need to create a file called _pi_logged_in (e.g. with touch _pi_logged_in) before calling make for the first time. On a different machine, you don’t need to do this. You cannot use make capture or make get without these command-line tools.

If future updates of mega-cmd or the command-line Kieler Compiler kico.jar include relevant breaking changes, you might need to resolve those problems (work without mega-cmd or acquire a kico.jar from around March 2020).

Setting up the Pi for Better Real Time Performance

The program will stick its latency-sensitive threads for controlling the IO pins and for image processing to the cores 3 and 4 of the 4-core Raspberry Pi. To ensure best real-time performance, please configure the kernel parameter isolcpus=2,3 (they count from 0) to the file /boot/cmdline.txt and reboot. This will make sure that the operating system doesn’t schedule any tasks on those cores.

Building and Running on Any Other Linux Machine

The program can be built and partially run on other machines. There you potentially need to install all the requirements yourself. Just wiringPi is not needed. Also don’t use a machine where the current user is called pi. We’ve observed problems with a too old version for make, for example 4.2 should be new enough. Possibly python2 needs to be installed on your machine. Compared to a Raspberry Pi, compilation of opencv on your machine is going to be a lot more fun (way faster and no issues with too little RAM) Regarding versions, see the relevant section above, note that pkg-config is needed, too, in case you don’t have that already.

Running works like on the Pi. The main program will prompt you for the name (without extension) of a video file in the video folder. Make sure to get some file, at the time of this writing there is a file called 1 that you can get via make get as well as a few more named files that were created before we had the make capture command.

After choosing a video file, the program starts using the video input in “real time”; since things are not really real-time, the controller does not work too well. (Simulating time properly is not fundamentally a problem but something we didn’t finish implementing.) This means you can mostly ignore the debug-printouts about ticks happening. What you can do is test the image processing / ball detection. This works better than on the Pi, because we are providing controls for pausing or changing the playback speed.

If you recalibrate the coordinate system on the Pi, you might want to transfer the config file for the coordinate system (however there is nothing really depending on the coordinate transformation being correct inside the image detection anyways).