dennybritz
A3C with Continuous Action Spaces
Are there any plans on making a A3C implementation for Continuous Action Spaces? Im not able to find anyone who have done this although they mention it in the original paper.
I think I will do it when I have time in addition to TRPO algorithm
On Fri, Mar 10, 2017 at 9:40 PM, Henrik Larsson [email protected] wrote:
Are there any plans on making a A3C implementation for Continuous Action Spaces? Im not able to find anyone who have done this although they mention it in the original paper.
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/dennybritz/reinforcement-learning/issues/72, or mute the thread https://github.com/notifications/unsubscribe-auth/AYFlCS2B8cH4KcX3XjQeeEfm_h20lZ-sks5rkgmvgaJpZM4MaEWO .
I did one experiment about "Pendulum", it works good. But when I increase the action dimension, like 'BipedalWalker-v2', it's action variance jumps too high and not converges. I till cannot find the reason.
My Pendulum code can be found in here: https://github.com/MorvanZhou/tutorials/blob/master/Reinforcement_learning_TUT/10_A3C/A3C_continuous_action.py
@MorvanZhou The link to your Pendulum code gives 404... Could you update it?
