REPO NO LONGER IN USE! See my other repos: car_localisation_dev, car_localisation_server, car_localisation_tx2
NOTE: if you're doing a fresh install of everything, then install the latest Nvidia JetPack (full OS reflash required!!!). It means you probably won't need to do the weird TensorFlow install method below. Also will require the Zed SDK to be updated, and the Zed camera driver to be updated. It seems as though TensorFlow's repositories don't support the TX2, so follow these instructions if you're using the TX2: https://github.com/jetsonhacks/installTensorFlowJetsonTX Install the wheel file for Python 2.7 because that's what ROS supports Follow this to install TensorFlow Object Detection: https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/installation.md
But instead put a line in the form of this into your .bashrc file:
export PYTHONPATH=$PYTHONPATH:/home/[USERNAME]/git/models/research/:/home/[USERNAME]/git/models/research/slim/
You want to install the models repo into /home/[USERNAME]/git/
Install this: https://github.com/stevenlovegrove/Pangolin
Must install ROS. Also, I recommend you use "catkin build" rather than "catkin_make": http://catkin-tools.readthedocs.io/en/latest/installing.html
Other prereqs: sudo apt install ros-kinetic-ros-control ros-kinetic-ros-controllers ros-kinetic-gazebo-ros ros-kinetic-gazebo-ros-control
Change the CPU architecture at the bottom of car_localisation/orbslam2/CMakeLists.txt, otherwise you'll get "error adding symbols: File in wrong format" while compiling. If you want to add a new architecture, you'll need to download ORB-SLAM2 (https://github.com/raulmur/ORB_SLAM2), compile g2o and DBoW2 in the ThirdParty folder, and copy their .so files across.
Compiling them is as easy as going into their folder, entering "cmake ." then "make". Their .so files are found in their lib folders.
Driver is from here: https://github.com/stereolabs/zed-ros-wrapper
Ensure that you pick a release (or pull latest version) that supports the Zed SDK you have installed. Each Zed SDK version requires a specific CUDA version. Changing the version of CUDA requires flashing the relevant JetPack (Full OS reinstall!!!)