Skip to content

Latest commit

 

History

History
123 lines (74 loc) · 6.33 KB

Installation.md

File metadata and controls

123 lines (74 loc) · 6.33 KB
Raw

Installation

Ubuntu

Check this tutorial to install Ubuntu 16.04 as dual boot besides Windows 10. Choose the default option (install ubuntu alongside windows)

For installation of Ubuntu on MSI (team laptop), follow this tutorial. You might need to install latest Nvidia drivers. Check this link on how to do that. Check this link to find the latest Nvidia drivers.

Cuda

  • Check cuda verison by typing nvcc -- version on a terminal.
  • Download cuda drivers from Nvidia website and follow their installation instructions.
  • For command 4,

sudo apt-get install cuda

becomes

sudo apt-get install cuda-9-0

Tensorflow

We'll use Tensorflow GPU. First, install dependencies for it. Cuda toolkit and Nvidia drivers were already installed in sections above.

For cuDNN libraries , go to this link and select cuDNN version corresponding to cuda 9.0. Download the following:

  • cuDNN 7.x Runtime library for ubuntu 16.04 (deb)
  • cuDNN 7.x Developer library for ubuntu 16.04 (deb)
  • cuDNN 7.x Code Samples and user guide for ubuntu 16.04 (deb)
  • (optional) cuDNN developer guide, install guide and release notes.

While installing cuda command line tools:

sudo apt-get install cuda-command-line-tools

becomes

sudo apt install cuda-command-line-tools-9-0

Add this path to the LD_LIBRARY_PATH environmental variable. This basically means that you need to modify the .bashrc file (it's a hidden file in home folder of ubuntu). Go to home, press Ctrl + H to reveal hidden files, open .bashrc file and paste the following at the end:

export LD_LIBRARY_PATH=${LD_LIBRARY_PATH:+${LD_LIBRARY_PATH}:}/usr/local/cuda/extras/CUPTI/lib64

Now, we can finally install Tensorflow GPU. We recommend using Native pip method.

Install pip: sudo apt-get install python-pip

Install tensorflow (you might want same version as on Embedded computer/Jetson TX2): pip install tensorflow-gpu == 1.8

NOTE: Throughout course of development, it's highly recommended to stick to one version of python (either 2 or 3). We recommend installing everything for python 2 as it doesn't cause clashes with ROS down the line.

Tensorflow Object Detection API

Clone the Tensorflow models repository. It's awesome!

git clone https://github.com/tensorflow/models

Install all dependencies for object detection api.

Potential issues

export PYTHONPATH="${PYTHONPATH}:/home/ajinkya/models/research:/home/ajinkya/models/research/slim/"

ROS

Install ROS (Robot Operating System) and check this handy tutorial series to learn how to use it!

Running Instructions

Most of the software we need is installed. Now we'll look at how to run the code.

Create a Workspace

All our code would reside in what is called a catkin workspace. To create a workspace, follow this link

mkdir -p ~/PerceptionAndSlam_KTHFSDV1718

cd ~/PerceptionAndSlam_KTHFSDV1718/

catkin_make

Next

cd ~/PerceptionAndSlam_KTHFSDV1718/

git clone https://github.com/javirrs/PerceptionAndSlam_KTHFSDV1718.git

This creates a folder named 'PerceptionAndSlam_KTHFSDV1718' inside our catkin workspace (which for now is also called PerceptionAndSlam_KTHFSDV1718). Now rename thie inner folder from PerceptionAndSlam_KTHFSDV1718 to 'src' (source, as this is where our source code will reside).

More Installations!

We're not done yet! To run various sensors with our system we need to install ros-wrappers for them. A ros-wrapper is a ros package with pre-written code supplied by manufacturers/open-source heros. This makes our life easier when interfacing with the sensors. We'll see how. Install these wrappers by cloning their repos in the src folder.

There are a few additional wrappers needed. Just look at the error when running catkin_make (we'll get to it soon), google them and clone their github repos in src

Finally,

cd ~/PerceptionAndSlam_KTHFSDV1718/

catkin_make

Launching the Program

We're here. Congratulations on reaching this far! To launch the program we use something called a .launch file. You can imagine it as being a file containing launch sequence for a rocket-launch. All the launch files are to be kept in perc_slam_launch package. This package contains two folders, launch and rosbags. A rosbag is like a video file, but instead of a video, it has the ability to record all (or chosen) ros topics. These can then be played back and we can run innumerable experiments on them!

Now, before ros can recognize all packages in our workspace, we need to source our workspace to .bashrc.

cd ~/PerceptionAndSlam_KTHFSDV1718/

echo 'source devel/setup.bash' >> ~/.bashrc

source ~/.bashrc

Next time onwards, ros will know the contents of our workspace. Now finally, run

roslaunch perc_slam_launch play_rosbag.launch

to play one of the rosbags, or connect all the sensors and run

roslaunch perc_slam_launch perc_slam.launch

to run the system on live sensor data.