For The Win: population-trained visual team play

What changed

Scientific scope

Combined concurrent populations, two-tier optimization of internal reward schemes and multiscale recurrent representations to learn pixel-based QuakeIII capture-the-flag policies that cooperated and competed with unfamiliar teammates/opponents. Original2018v1 study, not a silent merge with expanded2019journal scenarios. Uses game-point event signals and evolved reward shaping, not pixels plus terminal win alone. Human/agent control resolution, reaction time and tagging accuracy differ; artificial accuracy reduction tests do not prove identical interfaces or universal human-level teamwork. PBT and earlier recurrent/RL ideas are inherited.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Original2018v1 study, not a silent merge with expanded2019journal scenarios. Uses game-point event signals and evolved reward shaping, not pixels plus terminal win alone. Human/agent control resolution, reaction time and tagging accuracy differ; artificial accuracy reduction tests do not prove identical interfaces or universal human-level teamwork. PBT and earlier recurrent/RL ideas are inherited.

Contributors

Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

Human-level performance in first-person multiplayer games with population-based deep reinforcement learning

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction