search

UoA-CARES / cares_reinforcement_learning

CARES Reinforcement Learning Package
11 stars 2 forks source link

Feature/normalization for mbrl #124

Closed qiaoting159753 closed 8 months ago

qiaoting159753 commented 8 months ago

Normalization of the states and prediction of state differences (next_state - curr_state) have been confirmed to help train a world model.

This pull request: memory_buffer.py:

  1. add a function to compute state statistics based on stored transitions.

helpers.py:

  1. add normalize and denormalize functions. (It is used in SimpleDynamic and EnsembleRewardWorld)