Building a Doom-Like World to Explore Agentic Systems
summary
Builds a controlled Doom-like testbed that mirrors game state into an agent layer so actions are observable attributable and reproducible; implements a three‑tier semantic mirror (structural at ~7 Hz, semantic lenses, and behavior consumers) with 1,000 agents synced to a 35 Hz tick loop. Uses GZDoom patched runtime, ZScript agents, a Rust MCP gateway, GPT‑4/OpenAI for slow vision lenses and a “mosquito dog” semantic cache to cut LM calls and cost.
AI-written summary · about
chapters (9)
- 0:00Agentic systems in games
- 2:00Game engines for agent environments
- 8:45Doom as agent lab
- 17:49Agent perception and mirroring
- 23:39Agent Smith scenario
- 30:03Policy guards and enforcement
- 34:39Game agents to production
- 40:05Q&A on agents
- 51:33Closing and future
related talks
05mArtificial Players in the Design Process: Developing an automated testing tool for game level ...
50mDomain-driven Game Design
01h 00mVR 401: Accelerating Quest Development with Agentic Workflows
59mGame AI 4N00bz
23mDOOM Postmortem: 30 Years of DOOM
59mPlaying Video Games One Frame at a Time
59mHow to build (un)believable worlds?
33mRun your favorite games everywhere with WASM: the BlazorDoom use case