EVO-RatSLAM: A Biologically-Inspired SLAM for Event-Based Visual Odometry

An event-based camera is a biologically-inspired vision sensor that captures changes in a scene by recording pixel-level events and delivering asynchronous event streams. Event-based cameras excel in capturing fast-moving scenes and high-dynamic range environments, scenarios where conventional cameras often struggle. These characteristics make them well-suited for advanced robotics and computer vision tasks. This paper introduces EVO-RatSLAM: a pioneering biologically-inspired SLAM technique specifically designed for computing visual odometry with event-based cameras. This is the first SLAM approach that utilizes event-based visual odometry to create consistent and robust maps of camera motion over large spatial areas and for extended periods of time. This method, inspired by the computational model of rodents' hippocampus, combines continuous attractor network models with event-based visual inputs to construct cognitive maps. Odometric information is obtained by aligning the events using a probabilistic model. We validate the performance of EVO-RatSLAM through multiple real-world datasets captured using the DVXplorer event-based camera in various scenarios.