Overview

Defender is a two-dimensional side-scrolling shooting game set on the surface of an unnamed planet. The player controls a space ship as it navigates the terrain, flying either to the left or right. A joystick controls the ship’s elevation, and five buttons control its horizontal direction and weapons. The object is to destroy alien invaders, while protecting astronauts on the landscape from abduction. Humans that are abducted return as mutants that attack the ship. Defeating the aliens allows the player to progress to the next level. Failing to protect the astronauts, however, causes the planet to explode and the level to become populated with mutants. Surviving the waves of mutants results in the restoration of the planet. Players are allotted three ships to progress through the game and are able to earn more by reaching certain scoring benchmarks. A ship is lost if it is hit by an enemy, or hit by an enemy projectile, or if a hyperspace jump goes wrong (as they randomly do). After exhausting all ships, the game ends.

Description from Wikipedia

State of the Art

Human Starts

Result Method Type Score from
399865.3 ApeX DQN DQN Distributed Prioritized Experience Replay
233021.5 A3C LSTM PG Asynchronous Methods for Deep Learning
56533.0 A3C FF (4 days) PG Asynchronous Methods for Deep Learning
47671.3 RainbowDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
36242.5 A3C FF (1 day) PG Asynchronous Methods for Deep Learning
34415.0 DuelingPERDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
33996.0 DuelingDDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
32246.0 DistributionalDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
27510.0 DDQN DQN Deep Reinforcement Learning with Double Q-learning
23666.5 PERDDQN (rank) DQN Prioritized Experience Replay
21093.5 PERDDQN (prop) DQN Prioritized Experience Replay
20634.0 PERDQN (rank) DQN Prioritized Experience Replay
18037.5 NoisyNetDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
15917.5 DQN2015 DQN Dueling Network Architectures for Deep Reinforcement Learning
14296.0 Human Human Deep Reinforcement Learning with Double Q-learning
1965.5 Random Random Deep Reinforcement Learning with Double Q-learning
-9001.0 DQN2015 DQN Asynchronous Methods for Deep Learning
-9001.0 GorilaDQN DQN Asynchronous Methods for Deep Learning

No-op Starts

Result Method Type Score from
411943.5 ApeX DQN DQN Distributed Prioritized Experience Replay
55492.0 NoisyNet-A3C PG Noisy Networks for Exploration
55105.0 RainbowDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
52917.0 A3C PG Noisy Networks for Exploration
47092.0 C51 Misc A Distributional Perspective on Reinforcement Learning
42253.0 NoisyNet-DuelingDQN DQN Noisy Networks for Exploration
42214.0 DuelingDDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
41324.5 DuelingPERDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
37896.8 DistributionalDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
37275.0 DuelingDQN DQN Noisy Networks for Exploration
35338.5 DDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
31286.5 PERDDQN (rank) DQN Dueling Network Architectures for Deep Reinforcement Learning
27951.5 DQfD Imitation Deep Q-Learning from Demonstrations
24558.8 DuelingPERDDQN DQN Deep Q-Learning from Demonstrations
24162.5 PERDDQN (prop) DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
23633.0 DQN2015 DQN Dueling Network Architectures for Deep Reinforcement Learning
23083.0 NoisyNetDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
20525.0 NoisyNet-DQN DQN Noisy Networks for Exploration
18688.9 Human Human Dueling Network Architectures for Deep Reinforcement Learning
18303.0 DQN DQN Noisy Networks for Exploration
11099.0 DDQN+PopArt DQN Learning values across many orders of magnitude
2874.5 Random Random Dueling Network Architectures for Deep Reinforcement Learning