Multi-Sensor Perception, Fusion, and Robust State Estimation for Autonomous Agricultural Vehicles in Complex Environments

Abstract

Autonomous agricultural vehicles (AAVs) are important technologies for advancing intelligent and precision agriculture. This paper reviews the latest research progress in multi-source fusion positioning and robust state estimation technologies for autonomous agricultural vehicles in complex agricultural environments, with a focus on analyzing the typical application requirements and technical characteristics in open farmland, dense orchards, enclosed greenhouses, and GNSS-degraded and GNSS-denied scenarios. This review summarizes the key environmental factors that affect the autonomous navigation performance of agricultural vehicles, including GNSS signal blockage and multipath effects, mechanical vibration, wheel and track slip caused by complex terrain, and interference from dynamic vegetation. Subsequently, classification and comparison were conducted based on multi-sensor hardware configurations and data fusion architectures. On this basis, robust state estimation methods for complex agricultural environments were further discussed, including Adaptive Kalman filtering, Factor Graph Optimization (FGO), learning-assisted multi-sensor fusion, and motion-constrained slip compensation methods. This review analyzes the practical applications of multi-source fusion positioning technologies in autonomous tractor path tracking, orchard robot navigation, greenhouse inspection, and intelligent agricultural operation systems, and analyzes the current main bottlenecks faced by these technologies. This review discusses future development directions and provides technical insights into the development of highly reliable autonomous agricultural equipment.