How severe does this issue affect your experience of using Ray?
- High: It blocks me to complete my task.
I have the same issue as 15748. So instead I am using RLModule.forward_inference() in place of Algorithm.compute_single_action(), as shown in the example here.
How can I access the MeanStdFilter (and presumably the entire connector) to preprocess the observations with the stored statistics before passing them to the RLModule?
Before, it was something like algorithm.workers.local_worker().filters['default_policy']
.
Thank you.