batch_rl
batch_rl copied to clipboard
Is there some errors with the actions in the Ms-Pacman dataset?
I downloaded $store$_action_ckpt.10.gz
and $store$_observation_ckpt.10.gz
in atari-replay-datasets/dqn/MsPacman/1/replay_logs
.
I found that action and observation do not match.
Specifically, the first 10 actions are array([2, 2, 2, 2, 6, 2, 2, 7, 7, 7], dtype=int32)
. And I found that the action definition here is the same as https://gymnasium.farama.org/environments/atari/ms_pacman/ , action 2 controls the character to move to the right.
I saved the first few frames of observation:
Why is the character moving upwards? Am I missing something? @agarwl @tangbotony @zhixuan-lin @google-admin