Let's install the Donkey software on your host PC. The only part where this differs between the three platforms, Mac OS X, Linux and Windows, is in the Miniconda software installation, so we'll get that out of the way first.
Go to a place where you want the stuff we'll be working on to be.
# e.g. on Linux or Mac
cd ~
# e.g. on Windows
cd c:\\Users\\yourUsername
Make a folder for your projects and cd to it:
mkdir projects
cd projects
Clone the Donkey repository using git:
git clone https://github.com/autorope/donkeycar
cd donkeycar
git checkout master
Create the Python Anaconda environment using the yml file from the repository:
# Windows
conda env create -f install\envs\windows.yml
# Mac
conda env create -f install\envs\mac.yml
# Linux/Ubuntu
conda env create -f install\envs\ubuntu.yml
# All three OS's
conda activate donkey
pip install -e .[pc]
Install the TensorFlow software requirements for Nvidia GPUs, which basically means:
Download and install NVIDIA drivers
Download and install the CUDA Toolkit
Download and install cuDNN (you should just copy the bin, lib and include folders from the zip to your cuda installation folder)
Download and install TensorRT 5.0 to improve latency and throughput for inference on some models (same as the above)
Which means installing PyCUDA (make sure nvcc is in your PATH):
pip install 'pycuda>=2017.1.1'
Downloading and installing TensorRT 5.0
Then you can finally:
conda install tensorflow-gpu==1.13.1