Deep Learning-Based Real-Time Precise Pose Estimation Using Differential Magnetic Signals in the Dual-Robot Processing System

Applying the magnetic positioning system (MPS) in the dual-robot processing helps measure the relative pose of the two robot end effectors when there is an aluminum-alloy workpiece between them. However, many MPS researchers focus on in vivo medical device positioning without high precision, real-time performance, and anti-interference capability. A deep learning-based real-time precise pose estimation method using differential magnetic signals is proposed in this paper. Firstly, a magnetic positioning sensor is designed with three magnetic measurement units, and the sensor outputs the differential magnetic inductions measured by the three units. Then, the actual magnet field is precisely modeled, and the model is identified with a few magnetic induction data through particle swarm optimization (PSO). Next, numerous magnetic induction difference data are generated by the precise model to train a deep neural network that can calculate the sensor’s pose in real-time. Finally, a magnetic positioning experiment is conducted using two 5-DOF micro-positioning platforms, showing that compared with other researchers, the proposed method has higher positioning accuracy and less calculation time. In a cylindrical space with 4 mm radius and 10 mm height at a distance of 30 mm from the magnet, the average position error is 14.42 μm, the average angle error is 1.53 mrad, and the calculation time is 4 ms.