Lucas Maes is a researcher at AMI Labs and a third-year PhD student at Mila, where he studies world models, JEPAs, and control. His research aims to improve learned models’ ability to represent, predict, and reason about how environments evolve over time.

His work explores several approaches to extending JEPA-style architectures, including gradient-based planning, hierarchical temporal abstraction, goal specification, and improved understanding of physical dynamics. He created LeWorldModel, a research project focused on world-model learning, as well as stable-worldmodel, an open-source implementation for developing and studying these systems.

Before joining AMI Labs, Maes worked as a Research Scientist at Meta in Montreal in 2026. He began his PhD in artificial intelligence at Mila in 2023 after earning a research-focused master’s degree in artificial intelligence from Université de Montréal. His trajectory places him within the emerging research community working to move AI beyond static pattern recognition toward systems that can model, predict, and plan within the physical world.