edbeeching
Adds the RobotFPS example
RobotFPS environment:
An arcade style FPS environment, trained using SB3 PPO, observations contain 3 stacked frames of data from a raycast sensor (extended to add the facing direction of any visible robot relative to the agent). Making this env took quite a bit of experimentation with different types of obs and rewards while observing the trained behavior.
Trained agent onnx inference:
https://github.com/user-attachments/assets/4e7ecd92-f2e9-49c3-8b75-1a1f67af7922
Todo:
- [x] Add readme
- [x] Add video to repository readme
Amazing example @Ivan-267 , I will have time to look at the code later this week.
@Ivan-267 Nice env! I love the idea. Just a suggestion, can you add a few maps instead of just one. I tried the FPS example, and found it quite monotonous. Other than that, amazing environment.
@Ivan-267 Nice env! I love the idea. Just a suggestion, can you add a few maps instead of just one. I tried the FPS example, and found it quite monotonous. Other than that, amazing environment.
Thanks for your feedback and the suggestion!
The example is meant for users to test and modify/reuse/build upon, so anyone could for example add their own map, different designs or additional gameplay mechanics. I definitely agree it would be fun to have more maps, but at least at the moment it's out of scope of the example (but could be considered for a future time).