Next-generation agricultural mechatronic systems: a review of multimodal perception fusion and air-ground collaborative edge networks

Abstract

As agricultural ecosystems transition toward automation, traditional monomodal sensors often face significant data acquisition bottlenecks in unstructured environments due to physical obstructions and meteorological interference. This paper presents a systematic review of multimodal information fusion (MMIF) technology, positioning it as the core engine for enhancing the operational accuracy and robustness of agricultural mechatronic systems. From the perspective of low-level hardware coupling between electronic sensors and mechanical actuators, this paper analyzes the collaborative mechanisms of LiDAR, thermal imaging, and high-resolution vision for complex tasks such as greenhouse harvesting and autonomous navigation in open fields. This paper focuses on integration solutions for low-level hardware, specifically the deep integration of inertial measurement units (IMUs), encoders, and multi-channel sensor arrays, using Kalman filtering algorithms to ensure robust attitude estimation and localization for heavy-duty agricultural robots operating on rugged terrain. Furthermore, this paper investigates the evolution of communication architectures from centralized processing toward edge intelligence (TinyML) and “air-ground integrated (UAV-UGV)” collaborative networks, which enable low-latency decision-making and cross-platform task scheduling. Finally, this paper discusses current engineering bottlenecks, including edge-computing power constraints and the “black-box” nature of artificial intelligence. It proposes a technical roadmap toward “cloud-edge-end” collaborative computing and explainable artificial intelligence (XAI), aiming to build a trustworthy and highly autonomous next-generation smart agricultural infrastructure.