|
At some point, I got it in my head that implementing machine learning algorithms was easy, so I should jump into the deep end and go for reinforcement learning. This was misguided to say the least
This project is my attempt at using Deep Q learning to pilot a basic autonomous drone. The drone, Monkeybot 9000, was given the goal of reaching the banana analog at the end of its platform.
Monkeybot was able to fall off the platform and propel itself using its two rear wheels while using the front wheel for balance. The entire simulation was created using the Unity engine and C# was used for the logic.
Included in this logic was a built from scratch feed-forward neural network. No machine learning C# libraries that availed themselves to me were compatible with Unity unless they took all the fun out of it, so I had
to implement everything myself.
By no fault of its own, Monkeybot was a total failure.
Did I have a terribly good grasp of how neural networks functioned at the time of programming it?
No.
Have I achieved a better understanding since then?
Yes!
Have I had the time to fix the network that I wrote to control Monkeybot?
Unfortunately, no again.
So, Monkeybot sits in a git repository waiting for his time to shine.
|
