search

chengxuxin / extreme-parkour

Train your parkour robot in less than 20 hours.
https://extreme-parkour.github.io
Other
472 stars 85 forks source link

Hidden States in RecurrentDepthBackbone #12

Closed TextZip closed 7 months ago

TextZip commented 7 months ago

Hi, I was wondering why the hidden_states are not being reset when the episode ends in the learn_vision pipeline from the on_policy_runner. I know that they are being detached but when the robot gets reset shouldn't it also have a fresh hidden state of none instead of the previous episode's hidden_state value ?

Thanks for the amazing repo..

chengxuxin commented 7 months ago

You are right. But we found this does not affect training but saves a lot of effort if you want to find all terminating envs and reset corresponding hidden states.