Atari
A classic arcade game environment widely used as a benchmark platform in reinforcement learning research. It served as the original testbed in which DQN (Deep Q-Network) was first validated, providing a standardized suite of games for evaluating agent performance.
Core Facts
Timeline (last 90 days)
Atari did not state in the announcement whether offline downloads or refunds would be offered after the software store closes.
The Atari VCS software store will officially close on February 1, 2027.
The C51 algorithm, proposed by Bellemare et al. in 2017, is the first algorithm to successfully apply distributional reinforcement learning to Atari games; it approximates the return distribution with 51 fixed support points and updates the distribution through a projection step.
The Atari benchmark includes dozens of pixel-level games, with high-dimensional image states that require convolutional networks to process, and is the standard test set for evaluating the generalization ability of deep RL algorithms.
Mobile parkour games have higher requirements for real-time performance and robustness than the Atari benchmark or Go.
The DQN system was able to reach or surpass human-level performance on 49 Atari games using only raw pixel input.
When evaluating RL frameworks, one should pay attention to whether reproducible performance benchmarks are provided, especially performance on Atari, MuJoCo, or large-model RLHF tasks.
Architectures such as IMPALA and Ape-X are representative implementations of the sampling-training decoupling pattern, achieving throughput improvements of tens of times compared to single-machine synchronous training on tasks such as Atari.
The key difference between Touhou Project and classic RL benchmarks such as Atari is that its challenge lies in the need to simultaneously track and predict the trajectories of a large number of fast-moving objects.
The complexity of Roguelike games is higher than that of tic-tac-toe and Atari mini-games, but lower than that of large-scale AAA games.
7 more timeline events
All Facts (19)
DQN is an algorithm proposed by DeepMind in 2013 that for the first time combined deep neural networks with Q-learning reinforcement learning, enabling AI to learn to play 49 Atari games.
90%UnverifiedDeepMind's DQN appeared on the cover of Nature in 2015, being the first to demonstrate that a single architecture could surpass human-level performance across multiple Atari games.
60%UnverifiedAtari did not state in the announcement whether offline downloads or refunds would be offered after the software store closes.
50%UnverifiedThe C51 algorithm, proposed by Bellemare et al. in 2017, is the first algorithm to successfully apply distributional reinforcement learning to Atari games; it approximates the return distribution with 51 fixed support points and updates the distribution through a projection step.
50%UnverifiedThe Atari benchmark includes dozens of pixel-level games, with high-dimensional image states that require convolutional networks to process, and is the standard test set for evaluating the generalization ability of deep RL algorithms.
50%UnverifiedMobile parkour games have higher requirements for real-time performance and robustness than the Atari benchmark or Go.
50%UnverifiedThe DQN system was able to reach or surpass human-level performance on 49 Atari games using only raw pixel input.
50%UnverifiedArchitectures such as IMPALA and Ape-X are representative implementations of the sampling-training decoupling pattern, achieving throughput improvements of tens of times compared to single-machine synchronous training on tasks such as Atari.
50%UnverifiedWhen evaluating RL frameworks, one should pay attention to whether reproducible performance benchmarks are provided, especially performance on Atari, MuJoCo, or large-model RLHF tasks.
50%UnverifiedThe key difference between Touhou Project and classic RL benchmarks such as Atari is that its challenge lies in the need to simultaneously track and predict the trajectories of a large number of fast-moving objects.
50%UnverifiedThe complexity of Roguelike games is higher than that of tic-tac-toe and Atari mini-games, but lower than that of large-scale AAA games.
50%UnverifiedOpenAI's Gym, DeepMind's Atari challenge, and Pokémon-series reinforcement learning experiments have long served as testing grounds in the field of reinforcement learning
50%UnverifiedCompared to classic Atari games, Pokémon-type games have more complex decision chains, involving multidimensional continuous decisions such as map exploration, catching Pokémon, developing attributes, building teams, and battle strategy
50%UnverifiedIn the late Cold War period, the licensing of Tetris changed hands multiple times among Nintendo, Sega, Atari, and numerous PC software companies, triggering fierce licensing disputes.
50%UnverifiedDeepMind's Atari AI system surpassed top human players in some games.
50%UnverifiedDeepMind's DQN agent achieved superhuman performance in multiple Atari games, but minor visual perturbations can cause performance to collapse.
50%UnverifiedCuLE only supports Atari games and does not cover the NES platform.
50%UnverifiedDQN was published in Nature in 2015, combining deep neural networks with Q-learning to achieve human-level performance on Atari games for the first time.
50%UnverifiedThe Atari VCS software store will officially close on February 1, 2027.
50%