Unity have been playing around with integrating machine learning (ML) frameworks into the engine for a while now. With a lot of success.
Early examples which use of reinforcement learning have included many examples
The latest release has shown the development of an interactive dog which has learnt to return a stick which the user has thrown.
One of the most interesting aspects of the demonstration is that Puppo has to learng to walk. Puppo is made using a ragdoll approach with its legs driven by joint motors. The ML system needs to learn how to activate the joint motors correctly to complete the task.
Overall, it is a really impressive display of ML working in Unity.
You can read more about the project here, including downloading the project files.