Team frAIburg - Audi Autonomous Driving Cup 2017
This repository contains team frAIburg's Code for the AudiCup 2017.
If you participate in AADC2018 we highly recommend to read the
short bullet list with notes on ADTF and to checkout
our wrapper library adtf_slim,
which simplifies pin creation and sending of ADTF native types
and also OpenCV types with no overhead.
Videos
Installation
Besides adtf, boost, and OpenCV, this code needs tensorflow (we tested with version 1.4), qpoaisis and eigen. We expect them to live in the folder ADTF/Lib, if you place them somewhere else make sure to adapt the CMakeLists.txt.
Tensorflow
To optimize tensorflow for the car, we recommend building it directly on the car.
In principal you can follow the official guide,
but instead of only building the pip package you also need to build the libtensorflow.so library
for the c api. The installation of the pip package is only necessary if you plan to execute the demo task
via the thrift server. The main autonomous mode uses the TF C-API to directly run the forward
pass from an existing buffer.
At the time of writing this, this could be achieved by
bazel build --config=opt --config=cuda //tensorflow/c:c_api //tensorflow/tensorflow/ //tensorflow/tools/pip_package:build_pip_package
.
Documentation:
Most documentation and additional explanations are contained in README.md files
in the respective folders.
Felix Plum, Philipp Jund, Markus Merlinger, Jan Bechtold, Lior Fuks.