search

Shuijing725 / CrowdNav_DSRNN

[ICRA 2021] Decentralized Structural-RNN for Robot Crowd Navigation with Deep Reinforcement Learning
https://sites.google.com/illinois.edu/crowdnav-dsrnn/home
MIT License
95 stars 23 forks source link

Algorithm Problem #19

Closed ShelyH closed 1 year ago

ShelyH commented 1 year ago

Hi,

I read your code. I have some questions about the code details. When using the ST-RNN, the masks are input into GRU as a parameter. Could you explain why you did this? What is the function of masks input？

Thanks!

Shuijing725 commented 1 year ago

Hi,

The masks are the same as "done" in the gym environment. If an episode is done, we use masks to clear the hidden states of GRU to all zeros. See this line.