Overview

Video Pinball is a loosesimulation of a pinball machine: ball shooter, flippers, bumpers and spinners. It includes a unique rollover bonus with an Atari Inc. logo on the playfield; hitting the logo four times results in an extra ball.

Most of the game play involves learning how to perform specific functions, such as launching the ball or activating the flippers, with the Atari joystick. Moving the joystick controller down pulls the pinball machine plunger back while pressing the joystick button shoots the ball into the playfield. The left and right flippers are activated by moving the joystick controller left or right. The ball can be nudged (as in nudging a table gently in real life) by holding down the joystick button and moving the controller in a particular direction.

Description from Wikipedia

State of the Art

Human Starts

Result Method Type Score from
873988.5 ApeX DQN DQN Distributed Prioritized Experience Replay
506817.2 RainbowDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
470310.5 A3C LSTM PG Asynchronous Methods for Deep Learning
455052.7 DistributionalDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
447408.6 DuelingPERDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
374886.9 PERDDQN (prop) DQN Prioritized Experience Replay
367823.7 DDQN DQN Deep Reinforcement Learning with Double Q-learning
331628.1 A3C FF (4 days) PG Asynchronous Methods for Deep Learning
295972.8 PERDDQN (rank) DQN Prioritized Experience Replay
241851.7 NoisyNetDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
214925.3 PERDQN (rank) DQN Prioritized Experience Replay
185852.6 A3C FF (1 day) PG Asynchronous Methods for Deep Learning
154414.1 DQN2015 DQN Dueling Network Architectures for Deep Reinforcement Learning
112093.37 GorilaDQN DQN Massively Parallel Methods for Deep Reinforcement Learning
110976.2 DuelingDDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
20452.0 Random Random Massively Parallel Methods for Deep Reinforcement Learning
20228.1 DQN2015 DQN Massively Parallel Methods for Deep Reinforcement Learning
15641.1 Human Human Massively Parallel Methods for Deep Reinforcement Learning

No-op Starts

Result Method Type Score from
949604.0 C51 Misc A Distributional Perspective on Reinforcement Learning
876503.0 DuelingDQN DQN Noisy Networks for Exploration
870954.0 NoisyNet-DuelingDQN DQN Noisy Networks for Exploration
565163.2 ApeX DQN DQN Distributed Prioritized Experience Replay
533936.5 RainbowDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
479197.0 DuelingPERDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
478646.7 DistributionalDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
429936.0 DQN DQN Noisy Networks for Exploration
406420.4 PERDDQN (prop) DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
322507.0 NoisyNet-DQN DQN Noisy Networks for Exploration
309941.9 DDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
294724.0 NoisyNet-A3C PG Noisy Networks for Exploration
282007.3 PER DQN Scalable trust-region method for deep reinforcement learning using Kronecker-factored approximation
282007.3 PERDDQN (rank) DQN Dueling Network Architectures for Deep Reinforcement Learning
270444.6 NoisyNetDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
229402.0 A3C PG Noisy Networks for Exploration
196760.4 DQN2015 DQN Dueling Network Architectures for Deep Reinforcement Learning
157550.21 GorilaDQN DQN Massively Parallel Methods for Deep Reinforcement Learning
101339.6 DuelingPERDDQN DQN Deep Q-Learning from Demonstrations
100496.6 ACKTR PG Scalable trust-region method for deep reinforcement learning using Kronecker-factored approximation
98209.5 DuelingDDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
70009.0 DDQN DQN Deep Reinforcement Learning with Double Q-learning
56287.0 DDQN+PopArt DQN Learning values across many orders of magnitude
42684 DQN2015 DQN Human-level control through deep reinforcement learning
19761 Contingency Misc Human-level control through deep reinforcement learning
19123.1 DQfD Imitation Deep Q-Learning from Demonstrations
17667.9 Human Human Dueling Network Architectures for Deep Reinforcement Learning
17298 Human Human Human-level control through deep reinforcement learning
16871 Linear Misc Human-level control through deep reinforcement learning
16257 Random Random Human-level control through deep reinforcement learning

Normal Starts

Result Method Type Score from
156225.6 ACER PG Proximal Policy Optimization Algorithms
37389.0 PPO PG Proximal Policy Optimization Algorithms
19735.9 A2C PG Proximal Policy Optimization Algorithms