Spaces:
Sleeping
Sleeping
serial_collector | |
header Serial Pipeline | |
title Serial Collector | |
|#99CCCC|serial_controller| | |
|#99CCFF|env_manager| | |
|#CCCCFF|policy| | |
|#FFCCCC|collector| | |
|#99CCCC|serial_controller| | |
start | |
:init collector, set its \nenv_manager and \ncollect_mode policy; | |
|#99CCFF|env_manager| | |
repeat | |
|#99CCFF|env_manager| | |
:return current obs; | |
|#CCCCFF|policy| | |
:<b>[network]</b> forward with obs; | |
|#99CCFF|env_manager| | |
:env step with action; | |
|#CCCCFF|policy| | |
:process transition; | |
|#FFCCCC|collector| | |
:save transition in cache; | |
if (for every env: \n env_i is done? <b>OR</b> cache is full?) then (yes) | |
if (is <b>sample_collector</b>?) then (yes) | |
note right: Only sample_collector will do so, \n episode_collector will not. | |
|#CCCCFF|policy| | |
:<b>[adder]</b> get train_sample from cache; | |
endif | |
|#FFCCCC|collector| | |
:save sample/episode for return; | |
if (env_i is done?) then (yes) | |
|#99CCFF|env_manager| | |
:env_i reset; | |
endif | |
endif | |
|#FFCCCC|collector| | |
repeat while (collected sample/episode is not enough?) | |
:return sample/episode; | |
stop | |