cs-5v5-bots
A CS:GO gameplay behavior dataset designed for imitation learning and multi-agent reinforcement learning research. It contains 195 million timesteps of state-action data across 10 agents, accompanied by who-hit-whom damage labels, and is released under the CC BY-NC 4.0 license.
Timeline (last 90 days)
The cs-5v5-bots dataset contains 195 million timesteps of 10-agent CS:GO state-action data with damage labels, and is freely available for research use.
The dataset is positioned by its publisher v0rt3ch as a clean, narrow-domain testbed rather than a faithful reproduction of human gameplay.
The dataset contains 9,515 bot matches, 218,768 rounds, and approximately 3,958 hours of gameplay.
Each agent in the dataset is sampled at 16 Hz.
Each timestep records three categories of information for each of the ten agents: actions (key inputs and mouse movements), state (posture, health, weapons, etc.), and who-hit-whom damage labels.
14% of the matches in the dataset include ten first-person perspective videos, aligned frame by frame.
The dataset's collection settings are expert bots, a single map, and pistols only.
The dataset's structured data is provided in Parquet format, and videos use the webdataset format.
The dataset uses the CC BY-NC 4.0 license, meaning attribution and non-commercial use.
The dataset is hosted on Hugging Face (v0rt3ch/cs-5v5-bots-aim_usp-3k), and the publisher provides full specifications at vor.tech.
All Facts (10)
The dataset is hosted on Hugging Face (v0rt3ch/cs-5v5-bots-aim_usp-3k), and the publisher provides full specifications at vor.tech.
50%UnverifiedThe dataset is positioned by its publisher v0rt3ch as a clean, narrow-domain testbed rather than a faithful reproduction of human gameplay.
50%UnverifiedThe dataset contains 9,515 bot matches, 218,768 rounds, and approximately 3,958 hours of gameplay.
50%UnverifiedEach agent in the dataset is sampled at 16 Hz.
50%UnverifiedEach timestep records three categories of information for each of the ten agents: actions (key inputs and mouse movements), state (posture, health, weapons, etc.), and who-hit-whom damage labels.
50%Unverified14% of the matches in the dataset include ten first-person perspective videos, aligned frame by frame.
50%UnverifiedThe dataset's collection settings are expert bots, a single map, and pistols only.
50%UnverifiedThe dataset's structured data is provided in Parquet format, and videos use the webdataset format.
50%UnverifiedThe dataset uses the CC BY-NC 4.0 license, meaning attribution and non-commercial use.
50%UnverifiedThe cs-5v5-bots dataset contains 195 million timesteps of 10-agent CS:GO state-action data with damage labels, and is freely available for research use.
50%