Overview

Up’n Down is a vertically scrolling game that employs a pseudo-3D perspective.[citation needed] The player controls a purple dune buggy that resembles a Volkswagen Beetle.[citation needed] The buggy moves forward along a single-lane path; pressing up or down on the joystick causes the buggy to speed up or slow down, pressing right or left causes the buggy to switch lanes at an intersection, and pressing the “jump” button causes the buggy to jump in the air. Jumping is required to avoid other cars on the road; the player can either jump all the way over them, or land on them for points.[citation needed]

To complete a round, the player must collect 10 colored flags by running over them with the buggy. If the player passes by a flag without picking it up, it will appear again later in the round. The roads feature inclines and descents that affect the buggy’s speed, and bridges that must be jumped. A player loses a turn whenever the buggy either collides with another vehicle without jumping on it, or jumps off the road and into the grass or water.

Description from Wikipedia

State of the Art

Human Starts

Result Method Type Score from
347912.2 ApeX DQN DQN Distributed Prioritized Experience Replay
105728.7 A3C LSTM PG Asynchronous Methods for Deep Learning
74705.7 A3C FF (4 days) PG Asynchronous Methods for Deep Learning
54525.4 A3C FF (1 day) PG Asynchronous Methods for Deep Learning
29443.7 PERDDQN (prop) DQN Prioritized Experience Replay
24759.2 DuelingDDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
22681.3 DuelingPERDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
19086.9 DDQN DQN Deep Reinforcement Learning with Double Q-learning
16626.5 PERDQN (rank) DQN Prioritized Experience Replay
12157.4 PERDDQN (rank) DQN Prioritized Experience Replay
9896.1 Human Human Massively Parallel Methods for Deep Reinforcement Learning
8747.67 GorilaDQN DQN Massively Parallel Methods for Deep Reinforcement Learning
8038.5 DQN2015 DQN Dueling Network Architectures for Deep Reinforcement Learning
3311.3 DQN2015 DQN Massively Parallel Methods for Deep Reinforcement Learning
707.2 Random Random Massively Parallel Methods for Deep Reinforcement Learning

No-op Starts

Result Method Type Score from
436665.8 ACKTR PG Scalable trust-region method for deep reinforcement learning using Kronecker-factored approximation
401884.3 ApeX DQN DQN Distributed Prioritized Experience Replay
103557.0 NoisyNet-A3C PG Noisy Networks for Exploration
93931.0 DuelingDQN DQN Noisy Networks for Exploration
89067.0 A3C PG Noisy Networks for Exploration
82555.0 DQfD Imitation Deep Q-Learning from Demonstrations
82138.5 DuelingPERDDQN DQN Deep Q-Learning from Demonstrations
61326.0 NoisyNet-DuelingDQN DQN Noisy Networks for Exploration
44939.6 DuelingDDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
33879.1 DuelingPERDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
22972.2 DDQN DQN Dueling Network Architectures for Deep Reinforcement Learning
22474.4 DDQN+PopArt DQN Learning values across many orders of magnitude
16769.9 DDQN DQN Deep Reinforcement Learning with Double Q-learning
16154.1 PER DQN Scalable trust-region method for deep reinforcement learning using Kronecker-factored approximation
16154.1 PERDDQN (rank) DQN Dueling Network Architectures for Deep Reinforcement Learning
15612.0 C51 Misc A Distributional Perspective on Reinforcement Learning
14255.0 NoisyNet-DQN DQN Noisy Networks for Exploration
12561.58 GorilaDQN DQN Massively Parallel Methods for Deep Reinforcement Learning
11693.2 Human Human Dueling Network Architectures for Deep Reinforcement Learning
11652.0 DQN DQN Noisy Networks for Exploration
9989.9 DQN2015 DQN Dueling Network Architectures for Deep Reinforcement Learning
9082 Human Human Human-level control through deep reinforcement learning
8456 DQN2015 DQN Human-level control through deep reinforcement learning
3533 Linear Misc Human-level control through deep reinforcement learning
2449 Contingency Misc Human-level control through deep reinforcement learning
533.4 Random Random Human-level control through deep reinforcement learning

Normal Starts

Result Method Type Score from
145051.4 ACER PG Proximal Policy Optimization Algorithms
95445.0 PPO PG Proximal Policy Optimization Algorithms
17369.8 A2C PG Proximal Policy Optimization Algorithms