Overview

In this game based, loosely, on the movie of the same name, you have to move through a maze (the halls of your ship in the manual), ala Pac-Man, collecting dots (destroying alien eggs).

If you collect the power dot (pulsar), you can kill any of the three aliens, for a short time. There are only three enemies in the maze at a time, there is a bonus item at times and only one power dot (pulsar) at a time. When you grab the pulsar, it will next appear in one of two other spots.

After you clear one level, you get a bonus game. You have to move up the screen to the prize at the top past several aliens, reminiscent of Freeway. You do not lose a man if you fail but you only have eight seconds to do it then you are off the the next, harder level.

Description from RetroGames

State of the Art

Human Starts

Result Method Type Score from
17731.5 ApeX DQN DQN Distributed Prioritized Experience Replay
6371.3 Human Human Massively Parallel Methods for Deep Reinforcement Learning
6022.9 RainbowDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
1997.5 DistributionalDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
1486.5 DuelingDDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
1334.7 PERDDQN (rank) DQN Prioritized Experience Replay
1191.0 PERDQN (rank) DQN Prioritized Experience Replay
1033.4 DDQN DQN Deep Reinforcement Learning with Double Q-learning
945.3 A3C LSTM PG Asynchronous Methods for Deep Learning
900.5 PERDDQN (prop) DQN Prioritized Experience Replay
823.7 DuelingPERDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
813.54 GorilaDQN DQN Massively Parallel Methods for Deep Reinforcement Learning
634.0 DQN2015 DQN Dueling Network Architectures for Deep Reinforcement Learning
570.2 DQN2015 DQN Massively Parallel Methods for Deep Reinforcement Learning
533.3 NoisyNetDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
518.4 A3C FF (4 days) PG Asynchronous Methods for Deep Learning
182.1 A3C FF (1 day) PG Asynchronous Methods for Deep Learning
128.3 Random Random Massively Parallel Methods for Deep Reinforcement Learning

No-op Starts

Result Method Type Score from
40804.9 ApeX DQN DQN Distributed Prioritized Experience Replay
9491.7 RainbowDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
7127.7 Human Human Dueling Network Architectures for Deep Reinforcement Learning
6875 Human Human Human-level control through deep reinforcement learning
6648.6 PERDDQN (prop) DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
6197.1 DuelingPERDDQN DQN Deep Q-Learning from Demonstrations
6163.0 DuelingDQN DQN Noisy Networks for Exploration
5778.0 NoisyNet-DuelingDQN DQN Noisy Networks for Exploration
4737.5 DQfD Imitation Deep Q-Learning from Demonstrations
4461.4 DuelingDDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
4203.8 PER DQN Scalable trust-region method for deep reinforcement learning using Kronecker-factored approximation
4203.8 PERDDQN (rank) DQN Dueling Network Architectures for Deep Reinforcement Learning
4055.8 DistributionalDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
3941.0 DuelingPERDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
3747.7 DDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
3213.5 DDQN+PopArt DQN Learning values across many orders of magnitude
3197.1 ACKTR PG Scalable trust-region method for deep reinforcement learning using Kronecker-factored approximation
3166.0 C51 Misc A Distributional Perspective on Reinforcement Learning
3069 DQN2015 DQN Human-level control through deep reinforcement learning
2907.3 DDQN DQN Deep Reinforcement Learning with Double Q-learning
2620.53 GorilaDQN DQN Massively Parallel Methods for Deep Reinforcement Learning
2404.0 DQN DQN Noisy Networks for Exploration
2403.0 NoisyNet-DQN DQN Noisy Networks for Exploration
2394.9 NoisyNetDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
2027.0 A3C PG Noisy Networks for Exploration
1899.0 NoisyNet-A3C PG Noisy Networks for Exploration
1620.0 DQN2015 DQN Dueling Network Architectures for Deep Reinforcement Learning
939.2 Linear Misc Human-level control through deep reinforcement learning
227.8 Random Random Human-level control through deep reinforcement learning
103.2 Contingency Misc Human-level control through deep reinforcement learning

Normal Starts

Result Method Type Score from
1850.3 PPO PG Proximal Policy Optimization Algorithms
1655.4 ACER PG Proximal Policy Optimization Algorithms
1141.7 A2C PG Proximal Policy Optimization Algorithms