Ignacio Alzugaray

I am a postdoctoral research fellow in Andrew Davison's Dyson Robotics Lab at Imperial College London (ICL). My research centers on Spatial AI, focusing on the intersection of machine learning and geometric computer vision.

My current research interests include:

• Efficient Neural SLAM & Novel Scene Representations: Developing hybrid methods that combine geometric approaches with AI-driven modules for real-time performance and advanced scene understanding.
• Information-Guided, Data-Efficient & Decentralized Spatial AI: Creating distributed systems and agents that reason about space, motion, and action while minimizing communication and computational resources.
• Self-Supervised and Explainable AI from Large Foundation Models: Leveraging knowledge from large pre-trained models to build interpretable and generalizable AI solutions in 3D Computer Vision and Robotics.

At ICL, I design pixel-parallelizable distributed algorithms for emerging sensor-processor architectures, i.e. cameras where each pixel has limited computational and communication capabilities. I have also worked on real-time visual SLAM using neural 3D scene representations and explored scene coordinate regression for these tasks. My postdoctoral work includes guiding other junior researchers on topics such as distributed multi-agent positioning, object pose estimation with registration of CAD to 3D neural scene representations, and robot policies based on LLMs and VLMs.

Prior to joining ICL, I was a postdoctoral researcher in Margarita Chli's Vision for Robotics Lab at ETH Zurich, where I also completed my PhD in 2021. My postdoctoral work at ETH Zurich includes mentoring researchers on continuous-time SLAM and distributed multi-agent SLAM. My doctoral work focused on asynchronous feature detection and tracking for event cameras, enabling their use in visual SLAM. During my PhD, I interned at Disney Research Zurich on AR/VR technologies for the PoseVR project, and at Facebook Reality Labs, working on visual relocalization using implicit scene representations.