I’m a about two-thirds the way through the Self-Driving Car Nanodegree and it’s been awesome to be learning at a very rapid pace. In the past four weeks, I turned in two separate projects (here and here). Both were definitely challenging projects, but have taught me a lot about the process of training a deep learning model. More importantly, this is only the tip of the iceberg when it comes to utilizing deep learning (or more specifically CNNs).
The Behavioral Learning project stood out to me more so than the Traffic Sign Recognition project. Most likely because it was a very interactive approach, where we trained a car to drive solo around a small video game track. Check out the video.mp4 file in my repo if you want to see the car slowly making its way around the test track.
Around two or three weeks ago, I read an awesome piece in The Atlantic that detailed Waymo’s self-driving car efforts. The section(s) about a virtual world that Waymo has created really stood out to me and had small parallels to the Behavioral Learning project. Due to the costly and somewhat limited nature of gathering real data from self-driving car prototypes, it can be incredibly powerful to train a deep learning model virtually. And while the Behavioral Learning project was very rudimentary in comparison to real world applications, it gave us a little bit of insight into this world.
So far, the Behavioral Learning project has been the highlight of this course. It will definitely be something that sticks with me as I move towards the job search at the end of this course. This is somewhat of a short update, but I’ll be back with another update after completing the next couple projects. Advanced lane detection (computer vision / image processing) and vehicle detection / tracking are on the horizon!