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.
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.
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.
World Model Ecosystem: Weight + USD Alignment
Interactive Euler-style map using estimated concept scores, not measured market data. Circle size = Ecosystem Weight, a heuristic proxy for maturity, adoption, research visibility, and industry relevance. USD Alignment estimates how strongly each domain appears to connect to OpenUSD, scene graphs, simulation-ready 3D data, or USD-based workflows. Physical overlap shows meaningful workflow or research overlap.
Score guide
Domain Details
Select a circle to see the closest governing body, standards organization, or official ecosystem links. Not every domain has a single authority.
How to read the map
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.