Any update on continuous action spaces

PufferAI / PufferLib

Simplifying reinforcement learning for complex game environments

https://pufferai.github.io/

MIT License

1.3k stars 60 forks source link

Any update on continuous action spaces #125

Open kinalmehta opened 2 weeks ago

kinalmehta commented 2 weeks ago

I am working with continuous action space environments and could really use some speed-ups with the vector environments.

I am flattening the observations and actions by default.

Open to making a contribution if I understand the challenges with continuous actions spaces

jsuarez5341 commented 1 week ago

Hi -- these should work as is right now. The only thing that doesn't work yet is mixing discrete and continuous action spaces. Our emulation layer will even do that flattening for you. What issues are you having? Happy to reply here, but will be even faster on discord.gg/puffer