Robotic arm target localization strategy based on adaptive noise factor extended Kalman filter for depth camera and 2D LiDAR data fusion

Abstract

Traditional robotic arm target localization mainly relies on depth cameras, but its localization accuracy is easily disturbed under dynamic or complex lighting environments. To solve this problem, this paper proposes an adaptive noise factor extended Kalman filter (AEKF) multi-sensor fusion method to effectively fuse the data from 2D LiDAR and depth camera. Different from the traditional EKF method that uses a fixed observation noise covariance matrix R, AEKF dynamically adjusts the state and observation noise covariance to improve the stability and fusion accuracy in nonlinear and high-noise environments. In our experiments, we use a dynamically hovering coaxial UAV in the air as a grasping target to verify the real-time performance and grasping accuracy of the system. In addition, the proposed AEKF method does not require a large amount of model training to achieve efficient sensor fusion and can adaptively optimize the weights of each sensor according to the real-time residuals. The experimental results show that the proposed AEKF-based fusion system improves target detection accuracy by approximately 30%, and grasping success rate by around 8%–10%, compared to traditional single-sensor or EKF-based methods.