| On 2023-06-09 at (Brussels Time) |
Abstract
We present a robust method for visually segmenting scenes into objects by integrating motion and appearance cues through interconnected estimators. Our approach offers real-time, probabilistic, and consistent object segmentation, making it well-suited for a variety of robotic tasks. We showcase its effectiveness in kinematic structure estimation, where we observe significant improvements in object segmentation and estimated kinematic joints. Additionally, we explore the intriguing connection between the computational mechanisms of our segmentation method and the potential implications for collective estimation, where each agent can be considered as a distinct modality. Short Bio of the Speaker
Vito Mengers is a doctoral researcher specializing in robotics and intelligent behavior. With a master's degree in Computer Engineering from TU Berlin, Vito's research focuses on implementing diverse models of intelligent behavior in robots to explore the intricate relationship between perception and action. Advised by Prof. Oliver Brock, this work is part of a larger research endeavor within the Science of Intelligence cluster of excellence.
