Abdelrhman Werby

I'm a PhD student at the Institute for Artificial Intelligence, University of Stuttgart, where I'm advised by Kai O. Arras in the Socially Intelligent Robotics group. I'm also a member of the International Max Planck Research School for Intelligent Systems (IMPRS-IS).

Before that, during my master's at the University of Freiburg, I worked with Abhinav Valada and Wolfram Burgard on 3D scene graphs for language-grounded navigation and on articulated object estimation.

Email  /  Scholar  /  Github  /  LinkedIn  /  X

Research

I'm interested in robot perception and 3D scene understanding. My work focuses on building structured, open-vocabulary representations of the world, such as 3D scene graphs, and on estimating how articulated objects like doors, drawers, and cabinets move, so that robots can navigate and interact in everyday environments.

QueryArt estimates the articulation parameters of an object part from a single RGB image and a 2D query point, then recovers metric scale from one depth measurement. On a mobile manipulator it succeeds in 70.2% of 57 real-world manipulation trials.

Decoding functional parts and articulation from the latents of a generative 3D model, giving structural cues that can initialize robotic perception and planning.

Hierarchical 3D scene graphs built around keyframes, so the scene can be described and queried at multiple levels of detail.

ArtiPoint combines deep point tracking with factor graph optimization to estimate articulated parts and their axes from egocentric RGB-D video. We also release Arti4D, the first egocentric in-the-wild dataset of scene-level articulated object interactions.

HOV-SG builds a floor / room / object hierarchy of open-vocabulary 3D scene graphs, which lets a robot follow language queries across multi-story buildings.

Teaching

Website template from Jon Barron.