Closed TachikakaMin closed 2 years ago
The line of the encoder you're showing is only applied to image observations, not the reward. You can choose which observation keys should be used via these config options:
encoder.mlp_keys: '.*'
encoder.cnn_keys: '.*'
decoder.mlp_keys: '.*'
decoder.cnn_keys: '.*'
But only images (rank 3 tensor) and vectors (rank 1 tensor) are supported. The reward is scalar (rank 0 tensor).
For Plan2Explore, in expl.py the Class Plan2Explore will have a world model.
And this model will be WorldModel which is the same as dreamerv2.
For worldmodel training, the code will encode all information include reward into encoder
But Plan2explore says there should not be env reward.