Overview

The player controls the last defenses of the City of Atlantis against the Gorgon invaders. The city has seven bases, which are vulnerable to attack. Three of these have firepower capabilities to destroy the Gorgon ships before they manage to fire death rays at one of the settlements. The gun bases have fixed cannons; the center base fires straight up, while the far left and far right bases fire diagonally upwards across the screen. The center cannon also creates a shield that protects the settlements from the death rays, so once the center cannon is destroyed, the remaining settlements become vulnerable to attack. The enemy ships pass back and forth from left to right four times before they enter firing range, giving an ample opportunity to blow them away. Lost bases can be regained by destroying enough Gorgon ships.

Description from Wikipedia

State of the Art

Human Starts

Result Method Type Score from
918714.5 ApeX DQN DQN Distributed Prioritized Experience Replay
911091.0 A3C FF (4 days) PG Asynchronous Methods for Deep Learning
875822.0 A3C LSTM PG Asynchronous Methods for Deep Learning
814684.0 RainbowDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
772392.0 A3C FF (1 day) PG Asynchronous Methods for Deep Learning
629166.5 GorilaDQN DQN Massively Parallel Methods for Deep Reinforcement Learning
593642.0 PERDDQN (prop) DQN Prioritized Experience Replay
445360.0 DuelingDDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
423252.0 DuelingPERDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
330647.0 PERDDQN (rank) DQN Prioritized Experience Replay
319688.0 DDQN DQN Deep Reinforcement Learning with Double Q-learning
303666.5 NoisyNetDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
292491.0 DQN2015 DQN Dueling Network Architectures for Deep Reinforcement Learning
289803.0 DistributionalDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
207526.0 PERDQN (rank) DQN Prioritized Experience Replay
76108.0 DQN2015 DQN Massively Parallel Methods for Deep Reinforcement Learning
26575.0 Human Human Massively Parallel Methods for Deep Reinforcement Learning
13463.0 Random Random Massively Parallel Methods for Deep Reinforcement Learning

No-op Starts

Result Method Type Score from
3433182.0 ACKTR PG Scalable trust-region method for deep reinforcement learning using Kronecker-factored approximation
972175.0 NoisyNet-DuelingDQN DQN Noisy Networks for Exploration
944497.5 ApeX DQN DQN Distributed Prioritized Experience Replay
923733.0 NoisyNet-DQN DQN Noisy Networks for Exploration
920213.9 DQfD Imitation Deep Q-Learning from Demonstrations
902742.0 DuelingDQN DQN Noisy Networks for Exploration
876000.0 DQN DQN Noisy Networks for Exploration
841075.0 C51 Misc A Distributional Perspective on Reinforcement Learning
826659.5 RainbowDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
465700.0 NoisyNet-A3C PG Noisy Networks for Exploration
427658.0 PERDDQN (prop) DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
422700.0 A3C PG Noisy Networks for Exploration
395762.0 DuelingPERDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
382572.0 DuelingDDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
357324.0 PER DQN Scalable trust-region method for deep reinforcement learning using Kronecker-factored approximation
357324.0 PERDDQN (rank) DQN Dueling Network Architectures for Deep Reinforcement Learning
340076.0 DDQN+PopArt DQN Learning values across many orders of magnitude
329010.0 NoisyNetDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
303374.8 DuelingPERDDQN DQN Deep Q-Learning from Demonstrations
279987.0 DQN2015 DQN Dueling Network Architectures for Deep Reinforcement Learning
273895.0 DistributionalDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
106056.0 DDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
100069.16 GorilaDQN DQN Massively Parallel Methods for Deep Reinforcement Learning
85950.0 DQN2015 DQN Massively Parallel Methods for Deep Reinforcement Learning
85641 DQN2015 DQN Human-level control through deep reinforcement learning
64758.0 DDQN DQN Deep Reinforcement Learning with Double Q-learning
62687 Linear Misc Human-level control through deep reinforcement learning
29028 Human Human Human-level control through deep reinforcement learning
12850 Random Random Human-level control through deep reinforcement learning
12580.0 Random Random Noisy Networks for Exploration
852.9 Contingency Misc Human-level control through deep reinforcement learning

Normal Starts

Result Method Type Score from
2311815.0 PPO PG Proximal Policy Optimization Algorithms
1841376.0 ACER PG Proximal Policy Optimization Algorithms
729265.3 A2C PG Proximal Policy Optimization Algorithms