Creating an interesting problem for RL

Unity Engine
, ,
1

Dear scholar fellows, I have started creating a Pandemic Simulation with ml agents. My aim was to teach social distancing to agents. I have a github project called Pandemic Simulation and if anyone is interested can contribute.

From now on, I would like to hear your ideas about how can I extend this environment and create an actual interesting project from this point.

Let me introduce my settings:

The blue ones are the agents with nnModels and initially white ones are bots which pick random positions and go there. My agents learned in order to survive they need to protect their back =) Even though I am fascinating just by this I want to create a more sophisticated environment based on this. I am open to all recommendations and would like to hear what you think.

These are the colors and what do they represent:

2

From my experience, I noticed that the RL problem when getting complex (Mathematically Complex Problems) requires a massive amount of computing power to solve that is not available for practitioners at home. Therefore, I advise you to start simple and train. If it is trained very well then you can expand its complexity.

With that said, I recommend a simple environment with simple borders under the same layer/tag to mimic real-world borders and at the same time make it easier for RL to learn.

1 Like