Flockworld enables large-scale multi-agent video simulation and generation
FlockWorld: Toward Population-Scale Multi-Agent Video World Models through Artificial Life Simulation
Computer Vision and Pattern RecognitionMultiagent Systems
Summary
Most video models that predict what will happen next usually focus on a single character or a small group. The authors created FlockWorld, a computer environment that simulates large groups of agents moving together like flocks of birds, showing what each agent sees. They also made FlockDiT, a new method that can generate videos from these large groups, capturing multiple viewpoints at once. Their work provides a new way to study and generate complex multi-agent video scenes.
What this means in practice
- •For video game developers: Simulate and generate realistic multi-character video scenes to improve AI behaviors in large player populations.
- •For robotics system designers: Create multi-agent training environments with synchronized views for developing coordinated robotic behaviors.
Authors
Meryem Koksal, Yunzhe Wang, Volkan Ustun
Abstract
Most action-conditioned video world models remain limited to a single player, while existing multiplayer models support only small, fixed groups of agents. We introduce FlockWorld, an artificial-life environment built on Reynolds's Boids algorithm for studying population-scale multi-agent video world models. Specifically, FlockWorld renders synchronized egocentric crops, assigns distinct colors to preserve camera-agent identity across views, exposes rule-derived accelerations as actions, and lets the number of camera agents be configured independently of the size of the surrounding population. We further introduce FlockDiT, an action-conditioned latent diffusion transformer trained with flow matching on FlockWorld data to jointly generate synchronized egocentric videos for multiple agents. Our results suggest that joint generation benefits agent detection at density-matched settings and that diffusion forcing improves rollout fidelity. Together, FlockWorld and FlockDiT provide a controlled testbed for multi-agent video generation.