Advancing robotic systems with Ensemble and Modular Deep Learning: research idea and framework

Abstract

With the rapid growth of autonomy, robotics is undergoing a major transformation. Trends indicate a shift from manually programmed, predefined algorithms to adaptive, learning-based systems capable of navigating the complexities of dynamic and unknown environments. This paper investigates the integration of Deep Learning (DL) into robotic systems, focusing on the limitations of different DL algorithms in robotic applications. To overcome these challenges, we propose a novel research direction that addresses the current constraints and explores the potential of modular structures, such as ensemble and modular deep learning, to improve the adaptability, safety, and real-world applicability of robotic systems. The presented research framework outlines our future steps. The article concludes with possible solutions for the optimization of ensemble learning models.