Dynamic model has broad applications in motion planning, feedforward controller design, and disturbance observer design. Particularly, with the increasing application of model-based control in industrial robots, there has been a resurgence of research interest in accurate identification of dynamic models. However, on the one hand, most existing identification methods directly rely on least squares or weighted least squares (WLS), which suffer from outliers and could lead to physical infeasible solutions. On the other hand, nonlinearity of the friction model is seldom treated in a unified way with linear regression. Moreover, recent researches have shown that proper exciting trajectories are crucial to the identification accuracy, but few of previous works take measurement noise into consideration when optimizing the exciting trajectories. In this article, we propose an iterative approach which integrates WLS, iteratively reweighted least squares with linear matrix inequality constraints, and nonlinear friction models so that the above-mentioned issues can be properly solved altogether. Our research also reveals that performance can be improved by including priori knowledge of measurement noise in the optimization of exciting trajectories. The proposed approach is supported by experimental analysis of four different combinations within the framework on a 6-DoF industrial robot.

For a collaborative robot that shares the workspaces with humans, physical human-robot interactions are unavoidable. Therefore, it must be capable to detect collisions. Usual collision detection algorithms are based on the robot's dynamic model that frequently suffers from complex joint frictions. The uncertainty of commonly modeled frictions could lead to low sensitivity of the collision detection algorithm. This paper presents a novel collision detection scheme based on the robot's dynamic model that calculates the reaction force/torque at the bedplate. This dynamic model could avoid modeling joint frictions. To identify the dynamic model, a systematic procedure by measuring the force and torque at the robot's bedplate with a force/torque sensor is introduced. Moreover, a compensation method for the force/torque sensor signal coupling is developed. The proposed identification procedure has been demonstrated on a 6-DOF collaborative robot. Finally, collision detection experiments are conducted on a test-bed and humans. The results show that the proposed scheme is reliable and can effectively reduce the impact force with simple reaction strategies.

The accuracy of the force–torque sensors generally decreases with time under standard operating conditions. Therefore, periodic calibration for force–torque sensors is necessary. Specialized devices or subsidiary sensors are required during the traditional calibration process which bring in external measurement errors. In this article, an in situ calibration method is proposed to calibrate the force–torque sensor with a single standard mass to generate expected wrenches of the sensor during some arbitrary robot poses. Since the tilt of the mounting base of the robot can affect the direction of the gravity expressed with respect to the sensor frame, we first derive a model to express the relationship between the raw measurements and the expected wrenches under the influence of the tilted base, and then introduce subspaces and take advantage of projection and Rayleigh quotient to obtain the calibration matrix and angles of tilting simultaneously. The proposed algorithm is validated on simulation and the industrial automation (ATI) force–torque sensor. The results show that the proposed method is more cost-effective and reliable than existing technologies.

In servo systems, the dynamic characteristics may not only differ between axes but may also vary with moving directions for a single axis. The direction dependent characteristics would result in additional tracking or positioning error and degrade the performance of the system. In this paper, relay feedback tests are successfully applied to identify the dynamic characteristics in servo systems. A time-domain method is used to analyze the relay feedback other than the conventional describing function (DF) method. The time-domain method utilizes the same oscillation parameters (oscillation amplitude and half period) as the DF method for system identification. However, the time-domain method takes several advantages: First, the direction dependent characteristics of the system can be properly revealed; second, no approximation is made in this method, so that the exact expressions of the amplitudes and the periods of the limit cycles under relay feedback can be derived. A feedforward compensator is then designed using the estimated values of the system parameters. Simulation results show that the identification results through the time-domain method are more accurate than the DF method and are more robust under different relay parameters. Real time experiments show that the feedforward compensator designed by the proposed method compensates disturbances related to the direction and hence improves the tracking and positioning performance of the servo system.