Edged USLAM: Edge-Aware Event-Based SLAM
with Learning-Based Depth Priors

Şebnem Sarıözkan, Hürkan Şahin, Olaya Álvarez-Tuñón, Erdal Kayacan

Paderborn University | ICRA 2026

Code Full Dataset Paper

Abstract

Conventional visual SLAM algorithms often fail under rapid motion or low illumination. Edged USLAM introduces a hybrid visual-inertial system that extends Ultimate SLAM (USLAM) with an edge-aware frontend and a lightweight depth module. Our approach enhances event frames for robust feature tracking and provides region-of-interest based scene depth to improve motion compensation and scale consistency.

System Architecture

Overview of the Edged USLAM pipeline including Frontend, Backend, and Depth Module.

Edged USLAM Architecture

Aerial Dataset (TESTUDO Platform)

We provide a comprehensive event-based dataset collected via our TESTUDO drone platform, covering various challenges:

Drone
Illumination Sequences
Motion Sequences
Filtered Events
Unfiltered Events

Tracking Performance

Tracking Results

Comparison of Edged USLAM tracking performance under aggressive circular maneuvers.

BibTeX

@inproceedings{sariozkan2026edged,
  title={Edged USLAM: Edge-Aware Event-Based SLAM with Learning-Based Depth Priors},
  author={Sarıözkan, Şebnem and Şahin, Hürkan and Álvarez-Tuñón, Olaya and Kayacan, Erdal},
  booktitle={IEEE International Conference on Robotics and Automation (ICRA)},
  year={2026}
}