Achieving Deeper Player Interaction with Large-Scale Worlds
summary
Argues for combining high-detail local simulation with a large-scale persistent layer that runs deterministic paths and low-lod entities, spawning full NPCs only when within range. Cites Driver San Francisco traffic as thousands of cars moved cheaply by precomputed motion, contrasts typical 33 ms frame budgets and 20–30 active NPC caps with low-lod updates run every few seconds or even minutes and position ticks as infrequent as once per second. Recommends entity merging, multi-threaded independent updates and cloud services like SpatialOS/Hadean to keep world state persistent and responsive.
AI-written summary · about
chapters (16)
- 0:00Intro
- 0:40CONTENTS
- 6:18DEEPER INTERACTION
- 10:14HIGH QUALITY
- 11:28CPU LIMITATIONS
- 12:22DATA LIMITATIONS
- 13:29SIMULATION BUBBLE
- 16:06LARGE AREA
- 19:25LOW-LOD SIMULATION
- 20:48LOW FREQUENCY UPDATE
- 21:50ENTITY REDUCTION
- 22:53PLANNING AI
- 26:10SEMI-DETERMINISTIC SIMULATION
- 27:54THREADING
- 29:15CLOUD COMPUTE
- 30:12ONLINE WORLDS
related talks
23mForget everything you think you know about NPCs and world-building
43mOptimizing animations in Unreal Engine 4
50mEngine Tools for Performance on Open World Games
59mThe path to leveraging DOTS in production
15mGaming the Future of Architecture
39mBehind the Game: Strategies for Building the Tech That Powers Today’s Top Titles.
54mTen Lessons Learned
39mFantastic Bottlenecks and Where to Find Them