Embodied Agents in Dynamic Worlds

Speakers: Bryan Stout, John O'Brien

Track: Programming

Format: Lecture

Experience Level: Intermediate

Description: Until recently, pathing had become a solved problem. Creating, storing and using navigation information for a static world is fairly well-understood. With the latest round of hardware, however, developers are now building highly dynamic and changeable environments, which create new problems for the AI programmer.

In this lecture, we will cover successful approaches to navigating large numbers of agents through ever-changing environments. Topics will include dynamic search techniques, graph and mesh modification strategies, and on-the-fly navigation and avoidance techniques. The emphasis will be on covering the pros and cons of these approaches, including the performance implications and design/gameplay tradeoffs they entail.

Idea Takeaway: Attendees will leave with a clear understanding of the options and techniques available to them for moving agents through dynamic worlds. They will also have an appreciation for the performance and design tradeoffs implied by these choices.

Intended Audience: This talk is primarily aimed at programmers who will be involved in implementing navigation and pathfinding code. Lead programmers, designers and others interested in the tradeoffs and implications of highly dynamic worlds will also benefit. A familiarity with common pathfinding and navigation techniques such as A*, navigation meshes and steering behaviors will be assumed.