EgoCS-400K: An Egocentric Gameplay Dataset for World Models

arXiv 2026 57.5 benchmark

TLDR

Introduces EgoCS-400K, a large-scale egocentric gameplay dataset with aligned video-action-language trajectories for training interactive world models.

Reasoning

The paper's strength is providing a large, replay-grounded dataset with rich annotations (actions, states, events) that addresses a key data bottleneck for world models. A weakness is that it focuses solely on a single game (Counter-Strike), limiting visual diversity, and the abstract does not present any model evaluations or baseline results.

Read-first score

Read-first score 57.5, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 47.

Recency 6%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Citation impact 18%

92.9

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.92891518

Topical relevance 29%

67.1

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Methodology quality 18%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=dataset

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=dataset

Citation velocity 12%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FoundationFrontierBridge

Rank sensitivity

Stability: volatile; rank range: 454.

Keyword Scores

world model

interactive world model

video world model

world dynamics prediction

generative world model

world simulator

model-based reinforcement learning world model

Deep Analysis

Innovations

First large-scale egocentric gameplay dataset specifically designed for world models with temporally aligned video-action-language trajectories
Built from public professional Counter-Strike match demos, enabling replay-grounded parsing, rendering, and temporal alignment of player states, actions, camera motion, and events
Provides over 400,000 first-person videos and 10,000 hours of gameplay from 1,000+ matches and 40,000 rounds across 13 maps with 10 viewpoints per round
Supports multiple interactive visual modeling tasks: action-conditioned future prediction, state- and event-aware scene rollout, replay-grounded captioning, and agent egocentric action understanding
Serves as a practical bridge between passive web videos, controllable game simulation, and costly real-world embodied data

Methodology

The dataset is constructed by parsing public professional Counter-Strike (CS and CS2) match demos that preserve human gameplay trajectories. Player states, view directions, movements, keyboard/button inputs, view-angle changes, weapon usage, game events, and round-level context are extracted, and clean first-person videos are rendered from the same trajectories, enabling temporal alignment of video, action, and language.

Key Results

EgoCS-400K contains over 400,000 first-person videos and 10,000 hours of gameplay from more than 1,000 matches and 40,000 rounds, covering 13 maps and 10 player viewpoints per round.

Resources

Paper