Robust Multi-Sensor Fusion for Dynamic Robotic Grasping Using Fuzzy Adaptive Extended Kalman Filter

Abstract

Precise target localization and tracking are fundamental prerequisites for robotic arm grasping tasks in dynamic environments. Traditional single-sensor systems, such as depth cameras, are susceptible to environmental interference, particularly under varying illumination and complex backgrounds. To solve this problem, this paper proposes a novel lightweight multi-sensor fusion framework combining 2D LiDAR and depth camera data using a Fuzzy Adaptive Extended Kalman Filter (FA-EKF). Unlike standard EKF methods that use a fixed observation noise covariance matrix  , the proposed FA-EKF utilizes a fuzzy logic controller to dynamically adjust the noise covariance based on real-time measurement innovations. This allows the system to robustly handle non-linearities, sensor degradation, and time-varying noise. In the visual perception module, an improved YOLOv8 equipped with a Convolutional Block Attention Module (CBAM) is utilized to enhance feature extraction. Meanwhile, the LiDAR module extracts cylindrical targets using a robust RANSAC-based geometric fitting algorithm rather than traditional least-squares. Experimental validations performed on an AUBO-i5 robotic arm tracking a hovering coaxial drone demonstrate that the proposed FA-EKF method reduces the Root Mean Square Error (RMSE) to 0.0038m, and achieves a grasping success rate of 96% even under low-light conditions. The proposed system offers a real-time and highly robust solution for dynamic robotic grasping.