Is it possible to play against a trained bot from the keyboard using gym (Pong)?

Asked Nov 25 '19 at 22:24

Active Nov 25 '19 at 22:24

Viewed 303 times

I've been working on training a bot(DQN) on Pong using openai-gym and torch. After some success on the PongNoFrameskip-v4 environment, it occurred to me that it would be super nice if I would be able to play against this bot after training(and hopefully lose). Recently, I stumbled upon this keyboard agent, which seems to allow imitation training - training the bot from your plays. However, I haven't yet found a way to start an environment with my trained model as the right player(green) and play as the other player(brown).

Would really appreciate any advice regarding this topic. Thanks!

asked Nov 25 '19 at 22:24

John Szatmari

1

[This issue](https://github.com/openai/gym/issues/1134) (and links therein) might be informative. – a_guest Nov 25 '19 at 22:28
Thanks for the reply! From what I could see, the ````human_recorder.py```` would allow me to overwrite the controls of the left paddle while the right paddle would be controlled by the trained bot? – John Szatmari Nov 25 '19 at 22:53
1

Presumably yes, but I don't have any experience using it. – a_guest Nov 26 '19 at 00:17
Ok, I will try to adapt the script and make updates on progress – John Szatmari Nov 26 '19 at 08:38
1

Found this [issue](https://github.com/openai/gym/issues/1696) as well, hopefully there's more information here. – John Szatmari Nov 26 '19 at 08:51

Is it possible to play against a trained bot from the keyboard using gym (Pong)?

0 Answers0