Remi Cadene Hires for Embodied World-Models Team

Remi Cadene posted multiple times to recruit for an embodied world-models team, saying he will collaborate with Mustafa Shukor. What makes the posts notable is that they do more than announce openings: they also spell out a concrete technical bet for robotics and physical AI, namely scaling world models in latent space.

Core research direction

In Cadene’s hiring posts, the team says it places greater weight on scaling world models in latent space and applying them to physical AI. Framed this way, the emphasis is not on merely pushing scale along more familiar paths such as traditional robot control or end-to-end policies, but on expanding world-model capabilities themselves as the main lever.

Team and disclosed context

The recruitment posts say Cadene will work with Mustafa Shukor; the posts also identify Shukor as one of the authors associated with VL-JEPA and mention his collaboration background with Yann LeCun. Reposted commentary further describes this as a differentiated bet from many frontier robotics labs and says the effort is aimed at world models for robotics, including humanoid-robot latent world models.

Additional background from reposts

Reposted posts add that UMA is advancing this line of work with support from a SPRIND grant. They also sketch a broader research thread around scaling learning with minimal supervision and efficient data collection, citing directions such as self-supervision and third-person imitation. Based on the public posts in this cluster, the clearest external message remains: the team is hiring, focused on embodied world models, and treating latent-space world-model scaling as a central path for physical AI.

2026-07-17 ~ 2026-07-18 · 5 related posts