Overview

Asterix (Taz) was released by Atari in 1983 for the Atari 2600 and features the Looney Tunes character the Tasmanian Devil in a food frenzy. Within the game, Taz only appears as a tornado. The same game was released outside the United States featuring Asterix instead of Taz.

The gameplay is rather simple. The player guides Taz between the stage lines in order to eat hamburgers and avoid the dynamites. The game does not use any buttons and the difficulty increases by increasing the speed of the objects on screen. As the game progresses, the burgers may change into other edible or drinkable objects such as beer kegs, hot dogs, etc. There are not many sound effects in the game except a blipping sound when the player hits an edible object and another sound that resembles of explosion when the player hits dynamite.

Description from RetroGames

State of the Art

Human Starts

Result Method Type Score from
395599.5 DistributionalDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
364200.0 DuelingPERDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
283179.5 ApeX DQN DQN Distributed Prioritized Experience Replay
280114.0 RainbowDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
31907.5 PERDDQN (prop) DQN Prioritized Experience Replay
22484.5 PERDDQN (rank) DQN Prioritized Experience Replay
22140.5 A3C FF (4 days) PG Asynchronous Methods for Deep Learning
17244.5 A3C LSTM PG Asynchronous Methods for Deep Learning
16837.0 DDQN DQN Deep Reinforcement Learning with Double Q-learning
15840.0 DuelingDDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
9199.5 PERDQN (rank) DQN Prioritized Experience Replay
8277.3 NoisyNetDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
7536.0 Human Human Massively Parallel Methods for Deep Reinforcement Learning
6723.0 A3C FF (1 day) PG Asynchronous Methods for Deep Learning
3324.7 GorilaDQN DQN Massively Parallel Methods for Deep Reinforcement Learning
3170.5 DQN2015 DQN Dueling Network Architectures for Deep Reinforcement Learning
164.5 Random Random Massively Parallel Methods for Deep Reinforcement Learning
124.5 DQN2015 DQN Massively Parallel Methods for Deep Reinforcement Learning

No-op Starts

Result Method Type Score from
428200.3 RainbowDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
406211.0 C51 Misc A Distributional Perspective on Reinforcement Learning
400529.5 DistributionalDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
375080.0 DuelingPERDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
313305.0 ApeX DQN DQN Distributed Prioritized Experience Replay
41268.0 PERDDQN (prop) DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
32478.0 NoisyNet-A3C PG Noisy Networks for Exploration
31583.0 ACKTR PG Scalable trust-region method for deep reinforcement learning using Kronecker-factored approximation
31527.0 PER DQN Scalable trust-region method for deep reinforcement learning using Kronecker-factored approximation
31527.0 PERDDQN (rank) DQN Dueling Network Architectures for Deep Reinforcement Learning
28350.0 NoisyNet-DuelingDQN DQN Noisy Networks for Exploration
28188.0 DuelingDDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
18919.5 DDQN+PopArt DQN Learning values across many orders of magnitude
17356.5 DDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
15150.0 DDQN DQN Deep Reinforcement Learning with Double Q-learning
14328.0 NoisyNet-DQN DQN Noisy Networks for Exploration
12403.8 NoisyNetDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
11170.0 DuelingDQN DQN Noisy Networks for Exploration
8503 Human Human Human-level control through deep reinforcement learning
7566.2 DuelingPERDDQN DQN Deep Q-Learning from Demonstrations
6822.0 A3C PG Noisy Networks for Exploration
6433.33 GorilaDQN DQN Massively Parallel Methods for Deep Reinforcement Learning
6253.0 DQN DQN Noisy Networks for Exploration
6012 DQN2015 DQN Human-level control through deep reinforcement learning
5493.6 DQfD Imitation Deep Q-Learning from Demonstrations
4359.0 DQN2015 DQN Dueling Network Architectures for Deep Reinforcement Learning
1332 Contingency Misc Human-level control through deep reinforcement learning
987.3 Linear Misc Human-level control through deep reinforcement learning
210 Random Random Human-level control through deep reinforcement learning

Normal Starts

Result Method Type Score from
6801.2 ACER PG Proximal Policy Optimization Algorithms
4532.5 PPO PG Proximal Policy Optimization Algorithms
3176.3 A2C PG Proximal Policy Optimization Algorithms