Thanks for your issue! The agent is mainly trained and tested in the gym environments, especially the continuous control tasks. These environments provide standard interfaces so I just divided the continuous actions into discrete actions with equal length for convenience.
I believe this algorithm can adapt to tasks where different actions have different scales, with some slight adjustments in the codes. I'll consider updating my codes later. Thanks again!
Thanks for your issue! The agent is mainly trained and tested in the gym environments, especially the continuous control tasks. These environments provide standard interfaces so I just divided the continuous actions into discrete actions with equal length for convenience. I believe this algorithm can adapt to tasks where different actions have different scales, with some slight adjustments in the codes. I'll consider updating my codes later. Thanks again!