Mr. Tommy

For the final project in the Algorithmic Robotics and Motion Planning course given in Tel Aviv University by Prof. Dan Halperin, I wanted to implement some motion planning algorithms, such as PRM, RRT, RRT*, dRRT, dRRT* and some techniques such as hybridization.

During the course itself, I used the discopygal package, which certainly does raise the floor and ease the creation of such algorithms, and provides nice applications to design Scenes, i.e. environments with obstacles and robots which need to get from start to end, and a nice application to view Solvers, which are the algorithms that find a good route for the robots.

However, the package did have some caveats. For one, it is a nuisance to install, since it uses Conan and a the CGAL C++ project underneath the hood, and that requires some additional hacking for it to work. Another caveat of this approach is that the Python bindings of the C++ code are okay-ish, but not as friendly to work with as well-written Python code. The last caveat of the package that bothered me was that the tools were written in Qt, which is overall a fine choice, but I believe when developing such algorithms I want to be using Python and have everything very close to the code itself (and not in a separate app).

For this, I created the Tommy package. The package provides some interfaces written in Python with typing, which makes developing such algorithms easier. It should be noted that this comes with a performance cost of course, which I tried to cover by using Numba, but that didn’t turn out so well.

After creating that package, and completing my project, I wanted to think about the other issues I had with the package, and how I would go about solving them. For the scene editor, I created the Tommy VSCode Extension, which lets you work on the Scene’s JSON file, in a straightforward fashion, and see the scene in real time in a separate panel.

The current status of Tommy is alpha, with many lacking features and necessary performance improvements, but it does seem like a better approach to me nonetheless.