Overview

Skiing is a single player only game, in which the player uses the joystick to control the direction and speed of a stationary skier at the top of the screen, while the background graphics scroll upwards, thus giving the illusion the skier is moving. The player must avoid obstacles, such as trees and moguls. The game cartridge contains five variations each of two principal games.

In the downhill mode, the player’s goal is to reach the bottom of the ski course as rapidly as possible, while a timer records his relative success.

In the slalom mode, the player must similarly reach the end of the course as rapidly as he can, but must at the same time pass through a series of gates (indicated by a pair of closely spaced flagpoles). Each gate missed counts as a penalty against the player’s time.

Description from Wikipedia

State of the Art

Human Starts

Result Method Type Score from
-3686.6 Human Human 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
-10169.1 PERDDQN (rank) DQN Prioritized Experience Replay
-10852.8 PERDDQN (prop) DQN Prioritized Experience Replay
-10911.1 A3C FF (4 days) PG Asynchronous Methods for Deep Learning
-11359.3 ApeX DQN DQN Distributed Prioritized Experience Replay
-11490.4 DDQN DQN Deep Reinforcement Learning with Double Q-learning
-11685.8 RainbowDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
-11928.0 DuelingDDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
-12142.1 DQN2015 DQN Dueling Network Architectures for Deep Reinforcement Learning
-13247.7 DistributionalDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
-13700.0 A3C FF (1 day) PG Asynchronous Methods for Deep Learning
-13905.9 NoisyNetDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
-14863.8 A3C LSTM PG Asynchronous Methods for Deep Learning
-15287.4 Random Random Deep Reinforcement Learning with Double Q-learning
-18955.8 DuelingPERDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
-29404.3 PERDQN (rank) DQN Prioritized Experience Replay

No-op Starts

Result Method Type Score from
-4336.9 Human Human Dueling Network Architectures for Deep Reinforcement Learning
-7550.0 NoisyNet-DuelingDQN DQN Noisy Networks for Exploration
-7989.0 DuelingDQN DQN Noisy Networks for Exploration
-8857.4 DuelingDDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
-9021.8 DDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
-9900.5 PERDDQN (prop) DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
-9996.9 PERDDQN (rank) DQN Dueling Network Architectures for Deep Reinforcement Learning
-10789.9 ApeX DQN DQN Distributed Prioritized Experience Replay
-12630.0 DQN DQN Noisy Networks for Exploration
-12957.8 RainbowDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
-12972.0 A3C PG Noisy Networks for Exploration
-13062.3 DQN2015 DQN Dueling Network Architectures for Deep Reinforcement Learning
-13585.1 DDQN+PopArt DQN Learning values across many orders of magnitude
-13901.0 C51 Misc A Distributional Perspective on Reinforcement Learning
-14763.0 NoisyNet-DQN DQN Noisy Networks for Exploration
-14959.8 DistributionalDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
-15970.0 NoisyNet-A3C PG Noisy Networks for Exploration
-16307.3 NoisyNetDQN DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
-17098.1 Random Random Dueling Network Architectures for Deep Reinforcement Learning
-19949.9 DuelingPERDQN DQN Dueling Network Architectures for Deep Reinforcement Learning