Public Moneypit V for Victory source code

This repository contains the source code that ran the autonomous rover Moneypit V for Victory at the autonomous car race meet-up on August 12, 2017. The rover placed second.

The vision code runs on a Raspberry Pi 2, and lives inside the camcam directory. There is also a helper tool called showip which looks at the state of network interfaces, and every so often, tries to cycle them if they are not connected, as well as prints the IP address of the RPi on the display.

Talking to the roboclaw controllers and Dynamixel servos is done by a Teensy 3.2 microcontroller. The code for that lives in mpv_teensy.

Finally, this code uses some helpers to read/write images from the "stb" public repository (original source github.com/nothings/stb) which is included for convenience.

The Teensy code is built using Arduino for Raspberry Pi version 1.8.3, with the Teensyduino libraries installed from pjrc.com. The Raspberry Pi code is built using standard C++ and make. As is so often the case, this code is a living beast, mainly intended to do the one thing, with no real concern for discoverability or maintainability.

Some modules

I release it because there are some useful bits that could be re-purposed:

• The slimmed-down RaspiVid.c code, which can serve as a basis of other code that wants to read the RPi camera. It's less verbose and contains less special cases that you won't need than the original from Broadcom.

• The yuv code provides a convenient converter between YUV420 and RGB data.

• The settings module is simple yet highly useful for text-mode settings files.

• The pipeline and queue modules provide a simple way of sending data between multiple worker threads in a somewhat organized fashion.

• The detect_inner and project modules show how to do feature extraction for the yellow line on the track, and re-project it into a flat space that is then clustered. It's perhaps a useful illustration, but not intended to be highly reusable.

• The widget stucc in the gui module should be kept away from children and human beings with good taste, lest it rot out your eyes and gives you nightmares.

Training data

In the training_data directory, there are images collected while running the track; there is also a "training.csv" file that contains tuples of input filename, output steering, and output throttle response. This might be useful for someone who wants to train a neural network to clone the behavior of my rover :-)

Utilities

There are a few utilities:

• mkpng turns a raw RGB or YUV file into a .png file you can view with tools such as feh or whatever. Say ./mkpng {input.yuv} yuv to create an output file called raw.png; default size is 320x240.

• mkdetect runs the detction algorithm, given an input 320x240 image, and prints out what the algorithm thinks the throttle/steer response should be.

License

I place all this code in the public domain. I claim no responsibility for the correctness of the code -- in fact, it does have bugs, and should not be used in any kinds of systems where money, lives, animals, or anything else you care about are at stake. You take full responsibility for whatever you manage to do (or, more likely, fail to do) using this code.

2017 Jon Watte