World Models · Simulation · Synthetic Worlds

Where worlds are designed
and intelligence is formed.

World Model Research is an independent AI innovation studio focused on world models as the next foundation for AI and ML training. We build synthetic 2D/3D worlds, authored anomalies, and novel physics regimes where agents learn from experience, prediction, and interaction—not just text.

Founder & Research Lead: Joe Micallef · Blending AI, simulation, and hybrid 2D/3D worlds.
About

What is World Model Research?

World Model Research is dedicated to exploring how intelligent agents can learn from worlds instead of just data. As the limitations of large language models become clearer, world models are emerging as a successor architecture—systems that build internal representations of dynamics, causality, and affordances by interacting with synthetic environments.

We design and study 2D/3D hybrid simulation environments, authored anomalies, and novel physics regimes that challenge agents to understand causality, adapt to surprises, and imagine future states. This work blends machine learning, procedural and artist-driven world-building, animation, and creative technology.

Focus Areas

Research & Development

World-Model Architectures

Experimenting with Dreamer-style RSSM, JEPA-inspired predictive models, and other latent-dynamics architectures that learn to predict and imagine future states inside synthetic worlds.

Synthetic Physics Environments

Building worlds with both realistic and fantastical physics—gravity, flight, portals, spells— to test how agents generalize across regimes and learn causal structure.

Authored Anomalies

Designing hand-crafted, meaningful surprises (rule shifts, visual glitches, narrative triggers) that teach agents to detect change, adapt quickly, and reason beyond simple patterns.

Hybrid 2D/3D & World-Building Tools

Combining 3D simulation with 2D animation, billboards, and visual overlays to create worlds that are both expressive for humans and rich training substrates for machines.

Multi-Agent & Multi-View Data

Generating datasets with first-person views, minimaps, and schematic representations across multiple agents to support multi-view world-model training and representation learning.

Curriculum Worlds

Designing sequences of worlds with increasing complexity and shifting physics to study curriculum learning, cross-world transfer, and structural generalization.

Contact

Get in Touch

World Model Research is currently in an active exploratory and development phase, collaborating at the intersection of AI research, simulation, creative technology, and advanced visualization. For inquiries about collaboration, research opportunities, or project discussions, please reach out directly.

All inquiries:

Contact: Joe Micallef
Email: joe@worldmodelresearch.com
Phone: 310-985-1517

Joe combines decades of experience in animation, media, and technical education with a deep focus on world models, synthetic simulation, and hybrid 2D/3D environments for the next generation of AI.