Open maazashraf23 opened 5 months ago
I haven't worked with continuous action spaces before, but seems like there's a paper that tackles this problem (written by the authors of the original MuZero paper). I haven't read the paper myself yet.
I think you could also try quantizing the continuous action space into discrete actions, for example, converting a continuous action space of [-3, 3]
to a discrete action space of n
actions.
If n=3
, then your action space will be -3, 0, 3
If n=5
, then your action space will be -3, -1.5, 0, 1.5, 3
etc.
Hey Chiamp, how can i use this code to balance the continuous Single inverted pendulum environment with continuous action space? can you guide me little bit thanks.