Atari Freeway Environment

Overview

One or two players control chickens who can be made to run across a ten lane highway filled with traffic in an effort to “get to the other side.” Every time a chicken gets across a point is earned for that player. If hit by a car, a chicken is forced back either slightly, or pushed back to the bottom of the screen, depending on what difficulty the switch is set to. The winner of a two player game is the player who has scored the most points in the two minutes, sixteen seconds allotted. The chickens are only allowed to move up or down. A cluck sound is heard when a chicken is struck by a car.

Description from Wikipedia

Performances of RL Agents

We list various reinforcement learning algorithms that were tested in this environment. These results are from RL Database. If this page was helpful, please consider giving a star!

Star

Human Starts

Result Algorithm Source
29.1 Rainbow Rainbow: Combining Improvements in Deep Reinforcement Learning
28.9 Prioritized DDQN (rank, tuned) Prioritized Experience Replay
28.8 DDQN (tuned) Deep Reinforcement Learning with Double Q-learning
28.8 Distributional DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
28.4 Prioritized DQN (rank) Prioritized Experience Replay
28.2 PDD DQN Dueling Network Architectures for Deep Reinforcement Learning
27.9 Prioritized DDQN (prop, tuned) Prioritized Experience Replay
26.3 DDQN Deep Reinforcement Learning with Double Q-learning
25.8 DQN Massively Parallel Methods for Deep Reinforcement Learning
25.6 Human Massively Parallel Methods for Deep Reinforcement Learning
10.16 Gorila DQN Massively Parallel Methods for Deep Reinforcement Learning
0.2 DuDQN Dueling Network Architectures for Deep Reinforcement Learning
0.1 A3C FF 1 day Asynchronous Methods for Deep Reinforcement Learning
0.1 A3C FF Asynchronous Methods for Deep Reinforcement Learning
0.1 A3C LSTM Asynchronous Methods for Deep Reinforcement Learning
0.1 Random Massively Parallel Methods for Deep Reinforcement Learning

No-op Starts

Result Algorithm Source
34 DuDQN Noisy Networks for Exploration
34 NoisyNet DuDQN Noisy Networks for Exploration
34.0 QR-DQN-0 Distributional Reinforcement Learning with Quantile Regression
34.0 QR-DQN-1 Distributional Reinforcement Learning with Quantile Regression
34.0 Rainbow Rainbow: Combining Improvements in Deep Reinforcement Learning
34.0 IQN Implicit Quantile Networks for Distributional Reinforcement Learning
33.9 C51 A Distributional Perspective on Reinforcement Learning
33.6 Distributional DQN Rainbow: Combining Improvements in Deep Reinforcement Learning
33.3 DDQN A Distributional Perspective on Reinforcement Learning
33.0 PDD DQN Dueling Network Architectures for Deep Reinforcement Learning
32 NoisyNet DQN Noisy Networks for Exploration
31.8 DDQN Deep Reinforcement Learning with Double Q-learning
31.5 Reactor The Reactor: A fast and sample-efficient Actor-Critic agent for Reinforcement Learning
31.4 Reactor The Reactor: A fast and sample-efficient Actor-Critic agent for Reinforcement Learning
31 DQN Noisy Networks for Exploration
30.8 DQN A Distributional Perspective on Reinforcement Learning
30.3 DQN Human-level control through deep reinforcement learning
29.6 Human Dueling Network Architectures for Deep Reinforcement Learning
29.6 Human Human-level control through deep reinforcement learning
22.3 Reactor ND The Reactor: A fast and sample-efficient Actor-Critic agent for Reinforcement Learning
21.41 IMPALA (deep, multitask) IMPALA: Scalable Distributed Deep-RL with Importance Weighted Actor-Learner Architectures
19.7 Contingency Human-level control through deep reinforcement learning
19.1 Linear Human-level control through deep reinforcement learning
18 NoisyNet A3C Noisy Networks for Exploration
11.69 Gorila DQN Massively Parallel Methods for Deep Reinforcement Learning
0.0 ACKTR Scalable trust-region method for deep reinforcement learning using Kronecker-factored approximation
0.0 DuDQN Dueling Network Architectures for Deep Reinforcement Learning
0.0 Random Human-level control through deep reinforcement learning
0 A3C Noisy Networks for Exploration
0.0 IMPALA (shallow) IMPALA: Scalable Distributed Deep-RL with Importance Weighted Actor-Learner Architectures
0.0 IMPALA (deep) IMPALA: Scalable Distributed Deep-RL with Importance Weighted Actor-Learner Architectures

Normal Starts

Result Algorithm Source
32.5 PPO Proximal Policy Optimization Algorithm
0.0 A2C Proximal Policy Optimization Algorithm
0.0 ACER Proximal Policy Optimization Algorithm