Have you explored the impact of different normalization strategies for state/action.

[AaroniTurner]

Have you explored the impact of different normalization strategies for state/action (e.g., min-max, percentile-based like q99)?
If so, did you observe any noticeable differences in performance?
Thanks!