Multimodal Sensor Fusion for Autonomous Systems: Integrating Data from Various Sensors to Improve Decision-making in Autonomous Vehicles and Robotics
Michael Adewale Olufade
Department of Mathematics, Alcorn State University, Lorman, USA.
Emmanuel Aanu Bankole
Department of Mechanical, Aerospace and Nuclear Engineering, Rensselaer Polytechnic Institute, Troy, USA.
Olayinka Olubola Victor-Igun
Department of Computer Engineering, University of Benin, Benin, Nigeria.
Adekunle Junior *
Department of Computer Science, Usmanu Danfodiyo University, Sokoto, Nigeria.
*Author to whom correspondence should be addressed.
Abstract
Multimodal sensor fusion refers to the combination of data from various sensors to produce a more comprehensive and accurate understanding of the environment, enabling autonomous systems to make informed decisions. With the increasing adoption of autonomous vehicles and robotics, the need for robust and reliable sensor fusion techniques has become paramount. These systems must accurately interpret their environment, detect obstacles, and make rapid decisions to ensure safety and efficiency. Despite the numerous advantages, multimodal sensor fusion faces several challenges, including data synchronisation, computational complexity, and real-time processing demands. To address these demands, researchers are developing advanced algorithms and exploring machine learning techniques that optimise data processing. This paper presents a comprehensive review of multimodal sensor fusion techniques for autonomous systems, focusing on the integration of data from visual, acoustic, tactile, inertial, and environmental sensors to enhance decision-making in autonomous vehicles and robotics. By combining data from multiple sensors, multimodal sensor fusion enables autonomous systems to perceive their environment more accurately, improving obstacle detection, lane tracking, motion forecasting, and scene understanding. This review explores various sensor fusion techniques, including data-level, feature-level, and decision-level fusion, and discusses their applications in autonomous vehicles and robotics. Ultimately, this paper aims to contribute to the development of more robust and reliable autonomous systems, ultimately enabling safer and more efficient autonomous vehicles and robots. The paper also addresses challenges and limitations, such as sensor noise and uncertainty, data association, and computational complexity, and highlights future directions, including deep learning-based approaches and multi-agent sensor fusion. Computational complexity poses another challenge in sensor fusion. Integrating data from multiple sensors is computationally intensive, particularly when combining high-resolution inputs from lidar, radar, and visual sensors in real time. Moreover, real-time processing requirements present a limitation for sensor fusion in autonomous systems. Applications like autonomous driving or drone navigation require near-instantaneous processing of multimodal sensor data to make split-second decisions. The study concluded that Multimodal Sensor Fusion technology delivers an incredibly detailed and accurate picture of the environment, which is crucial for safe and efficient navigation.
Keywords: Multimodal sensor fusion, autonomous systems, autonomous vehicles, sensor integration