Question about the reward about rllab-curriculum

[florensacc]

Hi,
In our implementation we totally ignore the inner rewards - and in general any reward given by the environment. The rewards we consider are only the ones given by our custom wrapper environments: https://github.com/florensacc/rllab-curriculum/blob/master/curriculum/envs/goal_env.py#L171

[tldoan]

Hi,

Thank you for your prompt answer
I guess that link refers to the Goal GAN project?
I was playing with the reverse curriculum project
but yes I see how you implement your reward.

Thank you very much

[RushToNeverLand]

Hi,
In our implementation we totally ignore the inner rewards - and in general any reward given by the environment. The rewards we consider are only the ones given by our custom wrapper environments: https://github.com/florensacc/rllab-curriculum/blob/master/curriculum/envs/goal_env.py#L171

So in fact, your codes use the information (distance to goal) instead of sparse reward? For example: AntMaze, the reward in AntMaze is not sparse and your codes utilize the distance information?