Calibration of visual model for space manipulator with a hybrid LM–GA algorithm

• Build a hybrid LM–GA algorithm to optimize nonlinear equations with multiple peak values.
• Build a hybrid LM–GA camera calibration model of space manipulator.
• This model can improve locational accuracy of visual system.
• Design a novel measuring machine of space manipulator.
• This machine is used for dynamic orbital simulation and precision test.

A hybrid LM–GA algorithm is proposed to calibrate the camera system of space manipulator to improve its locational accuracy. This algorithm can dynamically fuse the Levenberg–Marqurdt (LM) algorithm and Genetic Algorithm (GA) together to minimize the error of nonlinear camera model. LM algorithm is called to optimize the initial camera parameters that are generated by genetic process previously. Iteration should be stopped if the optimized camera parameters meet the accuracy requirements. Otherwise, new populations are generated again by GA and optimized afresh by LM algorithm until the optimal solutions meet the accuracy requirements. A novel measuring machine of space manipulator is designed to on-orbit dynamic simulation and precision test. The camera system of space manipulator, calibrated by hybrid LM–GA algorithm, is used for locational precision test in this measuring instrument. The experimental results show that the mean composite errors are 0.074 mm for hybrid LM–GA camera calibration model, 1.098 mm for LM camera calibration model, and 1.202 mm for GA camera calibration model. Furthermore, the composite standard deviations are 0.103 mm for the hybrid LM–GA camera calibration model, 1.227 mm for LM camera calibration model, and 1.351 mm for GA camera calibration model. The accuracy of hybrid LM–GA camera calibration model is more than 10 times higher than that of other two methods. All in all, the hybrid LM–GA camera calibration model is superior to both the LM camera calibration model and GA camera calibration model.