Abstract: For some special manipulators such as the ones work at the space station, nuclear or some other unmanned environments, the overload, collision, vibration, temperature change or release of the internal stress would affect the structural parameters. And thus the operation precision might constantly decrease in long-term use. So the unattended manipulators should calibrate itself when they execute high precision operations or proceed self-maintenances. An on-line self-calibration method based on multi-markers is proposed. A camera fixed on the end of the manipulator is used to measure one to three positional makers, which forms an unstable multi-sensor eye-in-hand system. For the system has motion error and measurement error, a Gaussian motion (GM) method is proposed to form a feasible calibration model. The linear quadratic regulator (LQR) control and extended Kalman filter (EKF) are combined together to make the manipulator track the planned trajectories. In the same time, the Monte-Carlo method is proposed to form a high precision and stable signal acquisition when the visual system has measurement error and intermittent signal feedback. An automatic sampling process is adopted to select the optimal measurement points based on their variances. An expended DH model is proposed to approach 30 parameters in a time. Experiments are carried out on a manipulator in a life sciences glove box that will work at the Chinese space station to prove the efficiency and feasibility of the method proposed.

Key words: self-calibration, manipulator, Gaussian motion, visual measurement, system error