Research on Precise Perception and Error Compensation of End-Effector Pose for Robotic Arms Based on Multimodal Sensor Fusion

Precise end-effector pose perception remains a core bottleneck constraining intelligent robot performance enhancement. Existing fusion methods suffer from insufficient positioning accuracy due to neglecting nonlinear error coupling. This paper proposes a solution based on multimodal sensor fusion, constructing a sensor system that integrates binocular vision, a six-axis MEMS IMU, and a force tactile array. By leveraging the complementary capabilities of each sensor, the system achieves global precise measurement, dynamic continuous perception, and contact scenario feedback.Through temperature characteristic experiments, temperature-electromotive force model parameters are obtained. An error modeling approach combining deep learning and multi-source information fusion constructs a nonlinear coupling error model, separating multiple interference factors and establishing a digital twin real-time calibration system. Experimental validation demonstrates that this system effectively decouples multimodal heterogeneous errors and addresses various uncertainties, providing reliable perception support for sub-millimeter precision operations of robotic arms and the implementation of intelligent manufacturing technologies.