Cooperative Path Planning Implementation for a Robotic Manipulator and Magnetic Levitated Planar Actuator

Abstract

Magnetically levitated planar actuators (MLPAs) offer contactless motion, high precision, and kinematic flexibility, making them promising positioning platforms for industrial manufacturing applications such as continuous-seam welding, assembly, and inspection. However, their integration in cooperative tasks with robotic manipulators remains largely unexplored due to inherent controllability complexity and massive workspace. This thesis presents a comprehensive framework to enhance the control, operational range, and cooperative capabilities of a disc-magnet MLPA operating alongside a 4 degrees of freedom (DOF) robotic manipulator. To improve the overall standalone magnetic levitation system, the current implementation's control cycle is enhanced with a discrete Kalman Filter estimation and prediction sequence to provide smooth, optimal state estimation. To close the gap from uncertainties derived from device manufacturing, a mapping interpolation strategy is incorporated into the calibration workflow to increase the certainty of the calculated magnetic fields across the system. To expand the effective operational range, an enhanced magnetic force and torque dataset is generated to train a Dual Neural Network architecture, enabling extended actuation range across the entire levitation workspace without long loading times or high memory usage. The overall improvements and modifications are experimentally tested, obtaining similar results to the current implementation on the positional coordinates, but improved tracking metrics on the disc-magnet MLPA of ± 50% on all the tilting axes for a tested circular motion.

Building upon these core control enhancements, a cooperative framework is proposed in which the MLPA mover and the robotic arm execute a time-optimized planned path to complete complex trajectories with strict orientation constraints. The path-planning algorithm uses a geometrically modeled arm inverse kinematics and a filtered discrete arm workspace to generate a global search graph, resolving the system redundancies via Jacobian-based state configurations. To validate the flexibility of the proposed cooperative MLPA system, a trajectory-aligned baseline is developed in which the mover remains statically anchored during interaction. Experimental validation on a real MLPA system demonstrated that for standard geometries, the cooperative planner reduced completion time by almost 20% compared to the baseline, achieving positional and angular Root Mean Square (RMS) errors of 0.27 mm and 0.01°, respectively. Furthermore, evaluations on complex continuous-arc trajectories confirmed the critical necessity of cooperative mover–arm operation to synthesize feasible interaction solutions, conclusively demonstrating the potential of MLPA–robotic arm integration to expand the flexibility of kinematically limited industrial manufacturing lines.