My team and I are working on a newer project to train a car using computer vision techniques of our choosing to autonomously drive in a simulation we set up. This is a exciting field as many organizations are running to be the first to have a highly autonomous self-driving vehicle.

As we began researching possible avenues, we learned that this is indeed a very popular area of research.

So many blogs, tutorials, simulation environments, tools, and methods we able to found. The authors of these items ranged from students like us looking to expand our experience with some simple projects to highly professional organizations like Microsoft with a much more complex approach using their ‘AirSim’ package.

This has caused a bit of a roadblock because of how many options and ways there are to approach our project. While likely a good problem for our team to have, it gives us a lot of work to do to dig through. We must either find an approach that will be the most viable for our purposes or create our own Frankenstein of an approach using a variety of pieces of others.

Thankfully, I believe this is one of the coolest applications of machine learning (training in simulations that can be applied into real-world mediums) so I am very excited to dig in. I have not yet had the opportunity to experience much in this area, with my closest experience to this type of project being a workshop that introduced me into image recognition via a neural network with a small number of layers. However, what I might lack in experience I do my best to make up for with passion to dive into this field.