29 October 2021 in IEEE Transactions on Instrumentation and Measurement
This article is concerned with the identification of industrial robot dynamics parameters, and an identification method based on a nonlinear friction model and a least squares (LS) with symbiotic organisms search (SOS) algorithm is proposed. First, according to the friction characteristics of robot joints, for the nonlinear Tustin friction model, two parameters are introduced to describe the Stribeck characteristics of Coulomb friction and static friction. Then, the joint velocity trajectory that can offset the nonfriction torque during the forward and reverse rotation is designed, and friction parameters are estimated by using LS. Based on the optimal criterion constructed by Hadamard’s inequality, the excitation trajectory represented by the Fourier series is designed, and the combined inertial parameter set is estimated by using LS. Second, select the minimum inertial parameters and friction parameters (minimum dynamic parameters) as the center points to search the upper and lower bounds; design the fitness function with adjustable weights of each joint and use the SOS algorithm to estimate the minimum dynamic parameter set as a whole. Finally, a variety of parameter identification algorithms are applied to the Stäubli TX-90 robot under the same excitation trajectory. The experiment results show that the root-mean-square (rms) error of the joint torque estimated by the proposed method is the smallest. By executing the verification trajectory, the experiment results show that the rms error of the joint torque is small and the correlation coefficient is close to 1, thereby verifying the effectiveness of the estimated model for different motion trajectories.
10 February 2009 in Multidimensional Systems and Signal Processing
20 October 2012 in Circuits, Systems, and Signal Processing
20 July 2008 in Acta Automatica Sinica
26 September 2022 in IEEE Transactions on Instrumentation and Measurement
This study proposes an external force estimation method based on the error probability model and a sliding window variational adaptive Kalman filter (SWVAKF) for the problem of industrial robot external force estimation. A probability model with a normal distribution was provided, in which the covariance was linearly related to the joint velocity to address the uncertainty in the estimation error of the friction model. A normal distribution probability model was presented to address the time-varying characteristics of the contact external force when the robot interacts with the environment, in which the covariance of the derivative term of the external force was linearly related to the estimated external force. The robot interaction process was dynamically modeled using generalized momentum and external force as state variables, friction estimation deviation and external force derivative term as random interference, and generalized momentum with normally distributed noise as the measurement output. The parameters of the probability model of random interference items were determined by analyzing the experimental data of system identification, and the covariance related to joint velocity and external force was obtained. A set of adaptive scale factors related to the joint velocity and estimated force change rate when the robot interacts with the environment is proposed, and then the error covariance estimated based on the error probability model and the SWVAKF algorithm is fused by the adaptive scale factor, and a covariance of friction deviation and external force derivative term in the robot interaction process and an external force estimation method were provided. Experimental results reveal that the proposed method reduces the root-mean-square (RMS) error by 17.07% compared to the standard Kalman external force estimation method and has a better external force estimation effect.
13 May 2022 in IEEE Transactions on Neural Networks and Learning Systems
Industry 4.0 requires new production models to be more flexible and efficient, which means that robots should be capable of flexible skills to adapt to different production and processing tasks. Learning from demonstration (LfD) is considered as one of the promising ways for robots to obtain motion and manipulation skills from humans. In this article, a framework that enables a wheel mobile manipulator to learn skills from humans and complete the specified tasks in an unstructured environment is developed, including a high-level trajectory learning and a low-level trajectory tracking control. First, a modified dynamic movement primitives (DMPs) model is utilized to simultaneously learn the movement trajectories of a human operator's hand and body as reference trajectories for the mobile manipulator. Considering that the auxiliary model obtained by the nonlinear feedback is hard to accurately describe the behavior of mobile manipulator with the presence of uncertain parameters and disturbances, a novel model is established, and an unscented model predictive control (UMPC) strategy is then presented to solve the trajectory tracking control problem without violating the system constraints. Moreover, a sufficient condition guaranteeing the input to state practical stability (ISpS) of the system is obtained, and the upper bound of estimated error is also defined. Finally, the effectiveness of the proposed strategy is validated by three simulation experiments.