OpenAI Gym environments (Coders Strike Back)


#1

Hi everybody! I made a very simple OpenAi Gym environment inspired by engine of Coders Strike Back. It’s an old contest, but I think that it is one of the best game for learning and try new AI algorithms. This environment is very simple, it’s only a “single game mode”, i.e. you can control only one pod through a random set of checkpoints. Every episode is generated with a random set of maximum 15 checkpoints and it ends when the pod reach the last checkpoint. The reward mechanism is very simple, it just return a reward of -1 for every tick. I think that could be a good reward modelling for training a “Fast” pods, but if you have some other ideas let me know!

I would like to implement also the multi-pod version, with collisions. Contact me if you want to contribute in the development.


#2

@MadKnight getting his christmas present early this year.


#3

I modified:

  1. the reward model, now it give a reward of 100 for every reached checkpoint and -1 for every tick.
  2. Every episodes has a 32 random checkpoints
  3. Added pygame graphics :smiley:

#4

Nice idea! For the full game you can take a look at my sim engine implementation (in python).

I was planning to do something similar, though my idea was to expose python bindings to a pure C++ engine.


#5

Hi inoryy ! It’ great ! Can I use your code it to create a mini game with collision ? Do your engine equals the true game ? The C++ binding is a nice idea, it can be useful for a faster training of A.I.!


#6

Yep, feel free to use it. It should be correct, I’ve tested it with a tool when I wrote it.


#7

Ok, thank you