The end goal of MABDI is to be a software tool used for robotic applications, specifically environmental mapping. The current state-of-the-art algorithms that build a mesh map of the environment have an open-loop design that build mesh by simply by using using all of the available measurements from the senor, building a mesh, and then sequentially adding the mesh to a larger, global mesh. There are of course notable exceptions (reference Kinect Fusion and Kintinous) with their own separate cons. The main goal and idea of MABDI is to create a simple mechanism to first categorize the incoming measurements based on if they should be used to add to the global map of the environment. Essentially, closing the loop of the existing mapping algorithms and providing the system with a sense of introspection. My claim is that this will build maps that are more efficient (smaller) and require less computation to build.
MABDI uses CMake as the build process generator. If you are unfamiliar with CMake you should do some quick tutorials, but the idea is that you write a script for a process that looks at your source code and generates a build solution. By build solution, I mean a project that can be compiled and run. For example, if you are on Windows you will use CMake to build a Visual Studio project of if you are on a Mac you can generate an Xcode project or a Makefile. This can be very useful, for example, if you want your code to work in
the new Visual Studio version. The idea is that you change your CMake generator (which compiler to use) and your source code stays the same.
The complexity of the CMake system varies largely from project to project. MABDI's current system is very simple. It is meant to look for all the required third party libraries, generate the targets, and appropriatly link the targets to the libraries they may need.
Folder Structure:
- build - Not part of the repository, only mentioned in the .gitignore. User makes this folder. Use as a build area.
- src - upper directory for the MABDI code
- python - just has a Blender script to generate the .ply files I use
- tests - a sandbox area for playing with the code
- utils
- ply - folder with ply files
- environment - ply files for my simulation environment
- sensor - ply file for the kinect sensor
- ply - folder with ply files
A note about configuring, building, and installing third party libraries: Accomplishing this can be a true headache. This is because sometimes there are build errors and you have no idea why. Or, especially if you are new to CMake, you don't know how to configure a build. Also, sometimes you don't know what to configure. This is normal. You can do it. The key is to step back, be calm, perform a well thought google search, read what you feel is relevant (make sure not to make the mistake of trying to understand everything), and then try things. That is the best thing you can do. Try different things. Don't get stuck and frustrated. Keep track of your steps, and you'll be able to go back if something really doesn't work.
MABDI makes use of several third party libraries. I choose to do this for the whole "standing on the shoulders of giants" philosophy. It is assumed that you have built and installed the needed third party libraries before running CMake on MABDI. Current list of required libraries:
-
CMake 2.8.12 - Build generator
-
Qt 5.1 - GUIs
-
OpenGL- Graphics rendering
-
PCL 1.7.1 (Point Cloud Library) - Generating mesh from point sets
- Boost 1.55.0 - needed by many things including MABDI code
- Eigen 3.2.1 - Template library for matrix definition and methods
- Flann 1.8.4 - A library for performing fast approximate nearest neighbor searches.
- Qhull 2012.1 - Used in pcl surface
- Qt 4.8.5 - A gui development framework. Also used by Mabdi
- ICU 52 - Library providing unicode and globalization support for applications.
- VTK 5.10.0 - Visualization library
-
SDL version 2 - For joystick control
-
Doxygen - Documentation generator (optional)
-
LaTex - Needed by doxygen (optional)
-
Graphviz - Needed by doxygen (optional)
We are going to prefer dynamic link libraries because that seems to be the usual default when installing third party libraries. We will also prefer 32 bit builds of libraries to avoid incompatibilities.
Here are some links and notes about how to build and install the third party libraries on windows. Specifically, my platform is Windows 7 and I am building for 32-bit Visual Studio 2012.
Big picture notes: It is up to you where you want to place the third party
libraries. I chose to install them all in a folder named C:\3rdparty\ This
allowed me to think of all my tools being near the base of my file tree so that
I could easily incorporate them into new projects that I would have elsewhere.
TODO: Also, I added the binary folders of the libraries to my path, so this
kept the path names short.
Building each third party library can be quite different. In general, I followed these steps:
- Downloaded a .zip file of the latest stable version of the library to some folder in my work area. Then unzip the files.
- Ran CMake on the library and configured the project for my needs. Here is where you define static or dynamic, debug or release, paths to other needed thirdparty libraries, where to build the binaries, and other build properties.
- Moved the compiled library to folder with all third party libraries. Note, this does not mean all of the source code. Just the output, which can vary for each project, but often it's the contents of the folder where you told CMake to build the binaries.
Once again, the above is a very general guideline, below are some library specific notes. Not every library is mentioned, just the ones that I have some notes on.
Most of the libraries I needed to build were for the Point Cloud Library (PCL). I mostly followed this pcl tutorial in order to install the pcl specific library dependencies.
Notes on my environment:
- Installed Python 2.7
- Installed Perl
- Installed Ruby
- Installed jom
- Compiled and installed ICU 52
- Python, Perl, and Ruby I installed by using the installer provided by the project. This created the folders in my
C:\directory and added the paths to the binaries to my PATH variable. - I installed jom by unzipping the folder and adding the path to my PATH variable.
Notes on version choice. I would have prefered to use Qt 5 rather than Qt 4. This turned out to be a problem because I also wanted a full PCL build. PCL 1.7.0 has a problem with the VTK 6. I then switched to VTK 5 which then required me to switch to Qt 4.
- Boost 1.55.0
- Boost is needed by both PCL and MABDI. Boost can be thought of as a set of generally useful tools for C++. I am using version 1.55.
- I followed the Boost installation instructions
- I set the toolset (toolset=msvc) and left the rest as default. Run b2 --help to see default settings. This creates a dynamic library to be linked to.
- Qt 4.8.5
- Qt is needed by PCL, VTK, and MABDI.
- Go to the Qt download page. If your platform is supported, the easiest thing to do is to download the installer for your platform. This will install the binaries with no compiling necessary. If you are on Visual Studio, you will need the OpenGL supported build (not Angle) this is needed because of both VTK and MABDI.
- For me, I had to build the Qt project from source. Be warned, I had quite a few problems and also the build takes a few hours.
- For the most part I followed this write up from Qt
- I had to configure my environment for this install, so see "Notes on my environment" above.
configure -opensource -confirm-license -fast -debug-and-release -platform win32-msvc2012 -mp -opengl desktop -no-qt3support -no-xmlpatterns -no-multimedia -no-phonon -no-accessibility -no-openvg -no-webkit -no-script -no-scripttools -no-dbus -no-declarative- Then run
nmake
- lib/x64 of SDL2 had to be manually deleted. CMake insisted on finding the 64 bit library instead of the 32.
- CMAKE_PREFIX_PATH is hard coded for my particular setup
My memory is poor and if I don't write down all the ideas and facts swimming in my head it bogs down the main process a terrible amount. Result: many text files with a bunch of simple notes. Here is one for MABDI.