search

google-research / batch_rl

Offline Reinforcement Learning (aka Batch Reinforcement Learning) on Atari 2600 games
https://offline-rl.github.io/
Apache License 2.0
536 stars 75 forks source link

Is there some errors with the actions in the Ms-Pacman dataset? #40

Open hutchinsonian opened 5 months ago

hutchinsonian commented 5 months ago

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: obs0_a2 obs1_a2 obs2_a2 obs3_a2 obs4_a6 obs5_a2 obs6_a2

Why is the character moving upwards? Am I missing something? @agarwl @tangbotony @zhixuan-lin @google-admin

agarwl commented 5 months ago

I suppose this might be due to using the reduced action space (game dependent) in dopamine rather than full action space (18 actions) .

On Sat, Jun 15, 2024, 12:38 AM Zhang Jiahui @.***> wrote:

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: obs0_a2.png (view on web) https://github.com/google-research/batch_rl/assets/56984759/0ae93371-793a-4bc6-99bc-555e22f6f222 obs1_a2.png (view on web) https://github.com/google-research/batch_rl/assets/56984759/2c900aee-8752-4b6b-b575-a62963ddc488 obs2_a2.png (view on web) https://github.com/google-research/batch_rl/assets/56984759/cf00b07e-eb4f-4aa7-92e0-19c8c04e0628 obs3_a2.png (view on web) https://github.com/google-research/batch_rl/assets/56984759/9650ddb5-c8b0-4d57-8217-92ff2c10d06e obs4_a6.png (view on web) https://github.com/google-research/batch_rl/assets/56984759/0b767b62-2c66-4872-8fe6-d37f56fdc75b obs5_a2.png (view on web) https://github.com/google-research/batch_rl/assets/56984759/3fc84810-a163-4ec1-80e9-d13e5be603e3 obs6_a2.png (view on web) https://github.com/google-research/batch_rl/assets/56984759/fae11241-a90a-4ba8-84d1-9f14c03a6de9

Why is the character moving upwards? Am I missing something? @agarwl https://github.com/agarwl @tangbotony https://github.com/tangbotony @zhixuan-lin https://github.com/zhixuan-lin @google-admin https://github.com/google-admin

— Reply to this email directly, view it on GitHub https://github.com/google-research/batch_rl/issues/40, or unsubscribe https://github.com/notifications/unsubscribe-auth/AC4USYLFU3KR7DCZJBSYKLTZHPALBAVCNFSM6AAAAABJLLVPG2VHI2DSMVQWIX3LMV43ASLTON2WKOZSGM2TINJTHE4TSMI . You are receiving this because you were mentioned.Message ID: @.***>

hutchinsonian commented 5 months ago

I suppose this might be due to using the reduced action space (game dependent) in dopamine rather than full action space (18 actions) . … On Sat, Jun 15, 2024, 12:38 AM Zhang Jiahui @.> wrote: 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: obs0_a2.png (view on web) https://github.com/google-research/batch_rl/assets/56984759/0ae93371-793a-4bc6-99bc-555e22f6f222 obs1_a2.png (view on web) https://github.com/google-research/batch_rl/assets/56984759/2c900aee-8752-4b6b-b575-a62963ddc488 obs2_a2.png (view on web) https://github.com/google-research/batch_rl/assets/56984759/cf00b07e-eb4f-4aa7-92e0-19c8c04e0628 obs3_a2.png (view on web) https://github.com/google-research/batch_rl/assets/56984759/9650ddb5-c8b0-4d57-8217-92ff2c10d06e obs4_a6.png (view on web) https://github.com/google-research/batch_rl/assets/56984759/0b767b62-2c66-4872-8fe6-d37f56fdc75b obs5_a2.png (view on web) https://github.com/google-research/batch_rl/assets/56984759/3fc84810-a163-4ec1-80e9-d13e5be603e3 obs6_a2.png (view on web) https://github.com/google-research/batch_rl/assets/56984759/fae11241-a90a-4ba8-84d1-9f14c03a6de9 Why is the character moving upwards? Am I missing something? @agarwl https://github.com/agarwl @tangbotony https://github.com/tangbotony @zhixuan-lin https://github.com/zhixuan-lin @google-admin https://github.com/google-admin — Reply to this email directly, view it on GitHub <#40>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AC4USYLFU3KR7DCZJBSYKLTZHPALBAVCNFSM6AAAAABJLLVPG2VHI2DSMVQWIX3LMV43ASLTON2WKOZSGM2TINJTHE4TSMI . You are receiving this because you were mentioned.Message ID: @.>

In fact, in gymnasium Discrete(9), the action corresponding to action 2 is Right. image How to konw the actions defined in dopamine?

hutchinsonian commented 5 months ago

I also observed that the character is moving upwards when action 7 is executed at frames 30-32. I can't match these observations with the actions. This doesn't seem to fit any setting.

hutchinsonian commented 5 months ago

I found that these actions seem to be normal in the breakout environment. This also makes me wonder if there is some difference in the ms-pacman data?

I suppose this might be due to using the reduced action space (game dependent) in dopamine rather than full action space (18 actions) . … On Sat, Jun 15, 2024, 12:38 AM Zhang Jiahui @.> wrote: 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: obs0_a2.png (view on web) https://github.com/google-research/batch_rl/assets/56984759/0ae93371-793a-4bc6-99bc-555e22f6f222 obs1_a2.png (view on web) https://github.com/google-research/batch_rl/assets/56984759/2c900aee-8752-4b6b-b575-a62963ddc488 obs2_a2.png (view on web) https://github.com/google-research/batch_rl/assets/56984759/cf00b07e-eb4f-4aa7-92e0-19c8c04e0628 obs3_a2.png (view on web) https://github.com/google-research/batch_rl/assets/56984759/9650ddb5-c8b0-4d57-8217-92ff2c10d06e obs4_a6.png (view on web) https://github.com/google-research/batch_rl/assets/56984759/0b767b62-2c66-4872-8fe6-d37f56fdc75b obs5_a2.png (view on web) https://github.com/google-research/batch_rl/assets/56984759/3fc84810-a163-4ec1-80e9-d13e5be603e3 obs6_a2.png (view on web) https://github.com/google-research/batch_rl/assets/56984759/fae11241-a90a-4ba8-84d1-9f14c03a6de9 Why is the character moving upwards? Am I missing something? @agarwl https://github.com/agarwl @tangbotony https://github.com/tangbotony @zhixuan-lin https://github.com/zhixuan-lin @google-admin https://github.com/google-admin — Reply to this email directly, view it on GitHub <#40>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AC4USYLFU3KR7DCZJBSYKLTZHPALBAVCNFSM6AAAAABJLLVPG2VHI2DSMVQWIX3LMV43ASLTON2WKOZSGM2TINJTHE4TSMI . You are receiving this because you were mentioned.Message ID: @.>

agarwl commented 5 months ago

Based on the reduced action set, it does look like action 2 corresponds to moving up in MsPacMan. I got this from this colab

FULL_ACTION_SET = [
    'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
    'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
    'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
]

GAME_ACTION_SET = {
    'Alien': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'Amidar': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPFIRE', 'RIGHTFIRE',
        'LEFTFIRE', 'DOWNFIRE'
    ],
    'Assault': ['NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'RIGHTFIRE', 'LEFTFIRE'],
    'Asterix': [
        'NOOP', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT', 'DOWNRIGHT',
        'DOWNLEFT'
    ],
    'Atlantis': ['NOOP', 'FIRE', 'RIGHTFIRE', 'LEFTFIRE'],
    'BankHeist': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'BattleZone': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'BeamRider': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'UPRIGHT', 'UPLEFT', 'RIGHTFIRE',
        'LEFTFIRE'
    ],
    'Boxing': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'Breakout': ['NOOP', 'FIRE', 'RIGHT', 'LEFT'],
    'Carnival': ['NOOP', 'FIRE', 'RIGHT', 'LEFT', 'RIGHTFIRE', 'LEFTFIRE'],
    'Centipede': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'ChopperCommand': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'CrazyClimber': [
        'NOOP', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT', 'DOWNRIGHT',
        'DOWNLEFT'
    ],
    'DemonAttack': ['NOOP', 'FIRE', 'RIGHT', 'LEFT', 'RIGHTFIRE', 'LEFTFIRE'],
    'DoubleDunk': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'Enduro': [
        'NOOP', 'FIRE', 'RIGHT', 'LEFT', 'DOWN', 'DOWNRIGHT', 'DOWNLEFT',
        'RIGHTFIRE', 'LEFTFIRE'
    ],
    'FishingDerby': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'Freeway': ['NOOP', 'UP', 'DOWN'],
    'Frostbite': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'Gopher': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE'
    ],
    'Gravitar': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'Hero': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'IceHockey': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'Jamesbond': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'Kangaroo': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'Krull': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'KungFuMaster': [
        'NOOP', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'DOWNRIGHT', 'DOWNLEFT',
        'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE', 'UPRIGHTFIRE', 'UPLEFTFIRE',
        'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'MsPacman': [
        'NOOP', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT', 'DOWNRIGHT',
        'DOWNLEFT'
    ],
    'NameThisGame': ['NOOP', 'FIRE', 'RIGHT', 'LEFT', 'RIGHTFIRE', 'LEFTFIRE'],
    'Phoenix': [
        'NOOP', 'FIRE', 'RIGHT', 'LEFT', 'DOWN', 'RIGHTFIRE', 'LEFTFIRE',
        'DOWNFIRE'
    ],
    'Pong': ['NOOP', 'FIRE', 'RIGHT', 'LEFT', 'RIGHTFIRE', 'LEFTFIRE'],
    'Pooyan': ['NOOP', 'FIRE', 'UP', 'DOWN', 'UPFIRE', 'DOWNFIRE'],
    'Qbert': ['NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN'],
    'Riverraid': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    # 'RoadRunner': [
    #     'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
    #     'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
    #     'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    # ],
    'Robotank': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'Seaquest': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'SpaceInvaders': ['NOOP', 'FIRE', 'RIGHT', 'LEFT', 'RIGHTFIRE', 'LEFTFIRE'],
    'StarGunner': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'TimePilot': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPFIRE', 'RIGHTFIRE',
        'LEFTFIRE', 'DOWNFIRE'
    ],
    'UpNDown': ['NOOP', 'FIRE', 'UP', 'DOWN', 'UPFIRE', 'DOWNFIRE'],
    'VideoPinball': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPFIRE', 'RIGHTFIRE',
        'LEFTFIRE'
    ],
    'WizardOfWor': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPFIRE', 'RIGHTFIRE',
        'LEFTFIRE', 'DOWNFIRE'
    ],
    'YarsRevenge': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'Zaxxon': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
}
hutchinsonian commented 5 months ago

Based on the reduced action set, it does look like action 2 corresponds to moving up in MsPacMan. I got this from this colab

FULL_ACTION_SET = [
    'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
    'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
    'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
]

GAME_ACTION_SET = {
    'Alien': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'Amidar': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPFIRE', 'RIGHTFIRE',
        'LEFTFIRE', 'DOWNFIRE'
    ],
    'Assault': ['NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'RIGHTFIRE', 'LEFTFIRE'],
    'Asterix': [
        'NOOP', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT', 'DOWNRIGHT',
        'DOWNLEFT'
    ],
    'Atlantis': ['NOOP', 'FIRE', 'RIGHTFIRE', 'LEFTFIRE'],
    'BankHeist': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'BattleZone': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'BeamRider': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'UPRIGHT', 'UPLEFT', 'RIGHTFIRE',
        'LEFTFIRE'
    ],
    'Boxing': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'Breakout': ['NOOP', 'FIRE', 'RIGHT', 'LEFT'],
    'Carnival': ['NOOP', 'FIRE', 'RIGHT', 'LEFT', 'RIGHTFIRE', 'LEFTFIRE'],
    'Centipede': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'ChopperCommand': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'CrazyClimber': [
        'NOOP', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT', 'DOWNRIGHT',
        'DOWNLEFT'
    ],
    'DemonAttack': ['NOOP', 'FIRE', 'RIGHT', 'LEFT', 'RIGHTFIRE', 'LEFTFIRE'],
    'DoubleDunk': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'Enduro': [
        'NOOP', 'FIRE', 'RIGHT', 'LEFT', 'DOWN', 'DOWNRIGHT', 'DOWNLEFT',
        'RIGHTFIRE', 'LEFTFIRE'
    ],
    'FishingDerby': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'Freeway': ['NOOP', 'UP', 'DOWN'],
    'Frostbite': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'Gopher': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE'
    ],
    'Gravitar': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'Hero': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'IceHockey': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'Jamesbond': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'Kangaroo': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'Krull': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'KungFuMaster': [
        'NOOP', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'DOWNRIGHT', 'DOWNLEFT',
        'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE', 'UPRIGHTFIRE', 'UPLEFTFIRE',
        'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'MsPacman': [
        'NOOP', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT', 'DOWNRIGHT',
        'DOWNLEFT'
    ],
    'NameThisGame': ['NOOP', 'FIRE', 'RIGHT', 'LEFT', 'RIGHTFIRE', 'LEFTFIRE'],
    'Phoenix': [
        'NOOP', 'FIRE', 'RIGHT', 'LEFT', 'DOWN', 'RIGHTFIRE', 'LEFTFIRE',
        'DOWNFIRE'
    ],
    'Pong': ['NOOP', 'FIRE', 'RIGHT', 'LEFT', 'RIGHTFIRE', 'LEFTFIRE'],
    'Pooyan': ['NOOP', 'FIRE', 'UP', 'DOWN', 'UPFIRE', 'DOWNFIRE'],
    'Qbert': ['NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN'],
    'Riverraid': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    # 'RoadRunner': [
    #     'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
    #     'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
    #     'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    # ],
    'Robotank': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'Seaquest': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'SpaceInvaders': ['NOOP', 'FIRE', 'RIGHT', 'LEFT', 'RIGHTFIRE', 'LEFTFIRE'],
    'StarGunner': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'TimePilot': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPFIRE', 'RIGHTFIRE',
        'LEFTFIRE', 'DOWNFIRE'
    ],
    'UpNDown': ['NOOP', 'FIRE', 'UP', 'DOWN', 'UPFIRE', 'DOWNFIRE'],
    'VideoPinball': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPFIRE', 'RIGHTFIRE',
        'LEFTFIRE'
    ],
    'WizardOfWor': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPFIRE', 'RIGHTFIRE',
        'LEFTFIRE', 'DOWNFIRE'
    ],
    'YarsRevenge': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
    'Zaxxon': [
        'NOOP', 'FIRE', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT',
        'DOWNRIGHT', 'DOWNLEFT', 'UPFIRE', 'RIGHTFIRE', 'LEFTFIRE', 'DOWNFIRE',
        'UPRIGHTFIRE', 'UPLEFTFIRE', 'DOWNRIGHTFIRE', 'DOWNLEFTFIRE'
    ],
}

The action you mentioned 

'MsPacman': [
    'NOOP', 'UP', 'RIGHT', 'LEFT', 'DOWN', 'UPRIGHT', 'UPLEFT', 'DOWNRIGHT',
    'DOWNLEFT'
],

is the same as defined here https://gymnasium.farama.org/environments/atari/ms_pacman/. In MsPacman, action2 is RIGHT. So it does not correspond to observation

agarwl commented 4 months ago

Could it be one-based indexing? There's no reason for MsPacMan to be wrong specifically.