Showing posts from December, 2020

Building a Ball Balancing AI in Unity, Part 2

In Part 1 we looked at how to model and integrate an AI that balances a ball on a platform using neural networks and SmartEngine . To recap, this is the AI we are building: The final, trained AI in action When we last left, we had an integrated, functioning AI manipulating the torque on our platform, but we had not trained the AI. The result was that the AI failed miserably at its job because it was essentially giving random output. In this article, we're going to train our AI to produce the results we want. Let's begin! When you are ready to train an AI, you need to have a plan of attack. The first decision to make is what training methodology to use. SmartEngine provides three high level ways of training: gradient descent, reinforcement learning, and genetic training. Gradient descent is the tried-and-true method of training where the user supplies a set of i