A search and found zero-player game using ML-Agent. I was able to by using ML-Agent package and Code Monkey tutorials, Learn an Agent on a Custom map to catch goals.
agent and goal have four different positions that are randomly chosen for the positions of agent and goal, but each wayPoint position has a radius (0.5) that the position of agent and goal chosen in this radius of random waypoints position.
For learning this NN model for the agent I used a basic configuration but learned it on 3.5 million steps and the best mean reward was 0.993
You can see more information in the chart below
ml-agents: 0.29.0
ml-agents-envs: 0.28.0
Communicator API: 1.5.0
PyTorch: 1.7.0+cu110
Unity: 2020.3.14
1.I want to turn this into a chatch first game and pit the player against the agent
2.I want to make this agent smarter and increase its routing probability
I would be happy if you have any comments or suggestions for this project