Computer Setup

Contents

1. Install Ubuntu and Nvidia driver
2. Compiler Essentials
3. Install CUDA and CUDNN
4. Install ROS Kinetic
5. Tensorflow 1.10

1. Install Ubuntu and Nvidia driver

Note

The proposed packages requires ROS environment, CUDA support and a fully C++11-compliant compiler. Our codes have been fully tested under Ubuntu 16.04, ROS Kinetic. It should also be albe to run on newer version.

Tip

We recommend NVIDIA driver version 410 because for GeForce RTX 2080 the best supported CUDA version is 10.0 or above.

When the login screen appears, press Ctrl+Alt+F1. Enter your user name and password and then execute :

sudo add-apt-repository ppa:graphics-drivers/ppa
sudo apt update
sudo apt dist-upgrade
sudo apt install build-essential \
                 cmake \
                 pkg-config \
                 apt-utils \
                 git wget curl \
                 python-dev \
                 python-pip
sudo apt install nvidia-410
sudo reboot

2. Compiler Essentials

sudo apt install openssh-client \
                 openssh-server \
                 libgflags-dev \
                 libgoogle-glog-dev \
                 libprotobuf-dev
sudo apt install libtbb-dev \
                 libflann-dev \
                 gcc-multilib \
                 g++-multilib \
                 libboost-all-dev \
                 libglew-dev \
                 libjpeg8-dev \
                 libgtk2.0-dev \libv4l-dev \
                 qt5-default
sudo apt install libblas-dev \
                 liblapack-dev \
                 libsuitesparse-dev \
                 libatlas-base-dev \
                 libeigen3-dev \
                 gfortran \
                 libpcap-dev

3. Install CUDA and CUDNN

3.1 Prepare

Download CUDA 10.0 and CUDNN 7.5 in Share_UGV/softwares/cuda/cuda10.0 folder.

3.2 Install CUDA

sudo chmod +x cuda_10.0.130_410.48_linux.run
sudo ./cuda_10.0.130_410.48_linux.run

After above, accept EULA, do not install nvidia accellerated graphics driver (as already installed), as for the others, i.e. cuda-10.0_toolkit and cuda-10.0_Samples, you can choose to either install or not.

The default directory for the toolkit is: /usr/local/cuda-10.0.

3.3 CUDA envrionment Setup

echo "export PATH=$PATH:/usr/local/cuda/bin" >> ~/.bashrc

echo "export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib64:/usr/local/cuda/extras/CUPTI/lib64" >> ~/.bashrc

Then type source ~/.bashrc and nvcc -V. If you can see the CUDA version displayed, then CUDA is installed correctly.

3.4 Install CUDNN

tar -xvzf cudnn-10.0-linux-x64-v7.5.0.56
cd cudnn-10.0-linux-x64-v7.5.0.56/
sudo cp cuda/include/cudnn.h /usr/local/cuda/include
sudo cp cuda/lib64/libcudnn* /usr/local/cuda/lib64
sudo chmod a+r /usr/local/cuda/include/cudnn.h /usr/local/cuda/lib64/libcudnn*

4. Install ROS Kinetic

Follow the Installation page http://wiki.ros.org/kinetic/Installation/Ubuntu.

After ROS installation, install additional ROS packages:

sudo apt install python-catkin-tools \
                 ros-kinetic-cmake-modules \
                 protobuf-compiler \
                 ros-kinetic-serial \
                 ros-kinetic-octomap-mapping \
                 ros-kinetic-octomap-rviz-plugins \
                 ros-kinetic-rosbridge-suite

5. Tensorflow 1.10

Download the tensorflow file in Share_UGV/softwares/cuda/cuda10.0 folder.

Warning

This Tensorflow binary file is version 1.10 and only for Python2.7 and CUDA 10.0 GPU. If you want to install other versions or configurations of tensorflow, please go through TensorFlow Installation page.

Note

We suggest installing Tensorflow 1.10 under native python envrionment. If your computer has a conflicted version of Tensorflow, please install this version in a virtual envrionment.

Then execute:

pip install numpy --user
pip install tensorflow-1.10.1-cp27-cp27mu-linux_x86_64.whl --user