DonkeyCarGT: Donkey Car, C++ and You

This is the very early stage of an implementation of Donkey Car components in modern C++.

• Donkey Car is a self-driving car platform for hobby remote-control cars.
• Why? Good times.

Noteworthy

• Polymorphic architecture between Vehicle and Part (Parts are cameras, networks, computer vision components, etc).
• Smart-pointer, generic, polymorphic architecture for Part input and output data. The implementation of a Part may operate on a specific data type, e.g. a donkeycar::Image or may need to be generic to operate on multiple data types, e.g. NetworkPublisherMqtt<Pose>, NetworkPublisherMqtt<Image>. This is supported.
• Native, managed, thread support within Parts. A Part is a pthread that can be paused, resumed and joined with grace and care.
• Thread-safe, locked hashtable for Part I/O data passing, sharing and storage.

Status

• Architecture between Vehicle and Part
• Architecture for Part input and output data.
• Parts as managed threads
• Part I/O data storage and sharing
• Initial Support for Camera Parts: ImageListCamera (for iterating over Tub jpegs from disk) and a generic webcam WebCamera
• Throttle the Vehicle run loop to a specified frequency
• Significant unit-test coverage (always a WIP)
• .clang-format all code
• Autonomy support for CV-based line-following
• Autonomy support for Advanced CV-based line-following (WIP)
• Autonomy support for the Imitation Learning keras/TF-lite model (WIP)
• Support for Network Parts (mqtt and zmq with pub/sub semantics)
• Support for Tub_v2
• GPIO HW support
• ...

Building

Docker is effectively required, unless you're already working with:

• Tensorflow v2.13.0
• OpenCV >= 4.6.0

# Build the image
./docker/build-image.sh

# Edit the paths within this script, then run the container
./docker/run-image.sh

# Build donkeycarGT by making a build directory (i.e. `build/`), run `cmake` in that dir, and then use `make` to build the desired target.
mkdir build && cd build
cmake .. -DCMAKE_BUILD_TYPE=[Debug | Coverage | Release] -DTFLITE_ENABLE_XNNPACK=OFF
# or to modify the graph at runtime with the TFLite GPU delegate
# cmake .. -DCMAKE_BUILD_TYPE=[Debug | Coverage | Release] -DTFLITE_ENABLE_XNNPACK=OFF -DTFLITE_ENABLE_GPU=ON

make
./donkeyCarGT
make test      # Makes and runs the tests.
make coverage  # Generate a coverage report.
make doc       # Generate html documentation.