Humming Bird - A reinforcement learning experiment using ML-Agents in Unity

1- Goal:

To train an agent (the bird in this case) using a reinforcement-learning-based approach to learn how to drink nectar from flowers in the environment (the Unity scene).

2- Approach:

a) First, I started working on ML-Agents which is a machine learning library/package developed by Unity’s developers.
b) Then, by following ML-Agents and Unity documentation and a tutorial, I could set up a pre-constructed environment in which we have a bird and a scene containing some other objects like grass, trees, and rocks.

3- Unity Environment:

a) There are three types of object in the environment: 1) the agent (the bird), 2) the flowers, and 3) everything in the environment other than the flowers and the bird.
b) Using C# scripts in Unity, I defined specific behaviors for each of them during the training/inference stages.
c) By defining an action space, a state space, a reward term, and leveraging function APIs in Unity and ML-Agents, I could train the bird to perform the task described above.

Видео Humming Bird - A reinforcement learning experiment using ML-Agents in Unity канала AmirHossein Zamani