Welcome to the Robot Vision team at I3S (CNRS/Université Côte d'Azur). Situated at the intersection of robotics and artificial intelligence, the group focuses on bridging the gap between rigorous geometric perception and modern generative learning.

Our research addresses the fundamental challenge of robotic autonomy: enabling systems to perceive, represent, and interact with the physical world through persistent ego-centric spatial models. Moving beyond static mapping, the objective is to create "living," interaction-aware world models that support real-time operation and closed-loop action.

Our Research Axes

The team's work spans the full spectrum of spatial intelligence, organized around three converging pillars:

• Ego-centric Foundation Models & Generative AI:
  Current work defines emerging architectures for Ego-centric Foundation Models that fuse visual perception, language, and action. This includes the development of Generalisable Spherical World Models and Latent Inverse Graphics to simulate and predict environments using Neural Radiance Fields (NeRFs) and 3D Gaussian Splatting.


• Direct Methods & Dense Perception:
  Building on a strong legacy in pioneering Dense Direct Methods for visual SLAM, perception is formulated here through photometric consistency and continuous optimization. This approach ensures robustness in complex dynamic environments where feature-based methods fail.


• Embodied Action & Spatial Interaction:
  Spatial AI is viewed as a bridge between perception and interaction. Systems are validated on real-world platforms, ranging from humanoid robots in aircraft manufacturing (Airbus) to autonomous vehicles (Renault Group) and underwater multi-agent systems. A key focus is "Projective Inference"—giving robots the ability to understand the world from the perspective of their human collaborators.

Impact & Transfer

The team combines fundamental scientific innovation with high-impact industrial transfer. Most notably, we founded PIXMAP, a startup whose core dense mapping technology was acquired by a major US tech company and is now deployed in global-scale spatial computing platforms.

Today, major industrial partnerships continue to structure the agenda, working with leaders like Ampere (Renault Group) and Criteo to define the future of embodied AI.