Date of Award

Spring 5-19-2026

Access Restriction

Thesis

Degree Name

Master of Science

Department

Computer Science

School or College

Seaver College of Science and Engineering

First Advisor

Ray Toal

Abstract

Goal-Oriented Action Planning (GOAP) and Reynolds’ BOIDS flocking model are two foundational approaches to multi-agent AI in games. GOAP focuses on deliberate goal-driven decisions; BOIDS produces emergent group movement from local rules. Both have shipped in commercial games, but no published study has directly compared their performance and behavioral characteristics under controlled conditions. This thesis benchmarks three architectures in a Unity hack-and-slash demo: pure BOIDS, BOIDS directed by a GOAP leader, and per-agent GOAP with BOIDS movement. Each architecture implements the same nine-behavior set, runs against the same scripted player on the same arena, and is measured across six agent counts (25 to 800). The benchmark covers 1,080 trials across stress and combat modes, with medians and inter-quartile ranges on warm-up-trimmed metrics. BOIDS directed by a GOAP leader produces the highest combat decisiveness in this benchmark. It produces a player death in 40 to 60 percent of trials at every agent count tested, the highest damage rate against the player, and the fastest time to first hit, while staying within ten percent of the cheapest condition on per-frame CPU. PureBOIDS scales cleanly to large populations but never threatens the player on its own. Per-agent GOAP with BOIDS movement works at small agent counts but its combat effectiveness collapses as the swarm grows. For a fast-paced combat game with large enemy swarms, BOIDS directed by a GOAP leader is the strongest performer in this benchmark across both axes.

Download

Available for download on Wednesday, November 18, 2026

Share

COinS