No abstract

An auto-tuning method for a Delta robot's P/PI cascade motion controller using multi-objective optimization algorithm is proposed. The implemented control structure consists of two controllers: A feedforward controller based on a model of the inverse dynamics of the robot, and a cascade P/PI controller to compensate for unmodeled effects. The auto-tuning is achieved in the sense of optimizing the control parameters in three stages. In the first stage, the feedback control parameters are optimized after neglecting the feedforward control term. The goal is to minimize the position error in tracking an excitation trajectory, which is used as well to identify the dynamic model parameters in the second stage. After that, the feedforward compensation term is computed offline based on the desired trajectory. In the final stage, the P/PI parameters are optimized again after adding the feedforward controller. Experimental results on an industrial 3-dof Delta robot validates the efficiency of the proposed method.

This paper presents a sensitivity-based approach for optimal model design and identification of the dynamics of a state-of-the-art industrial robot considering process-related restrictions. The possibility of parameter excitation for subsequent identification of the model parameters is severely limited due to restrictions imposed by the process environment, especially the limited available workspace. Without sufficient parameter excitation, a satisfactory quality of the full model identification cannot be achieved, since non-excited parameters cannot be identified correctly. Furthermore, optimal excitation requires time-consuming calculations and distinct experiments during which the robot is not available for daily operation. It is therefore of interest to use process-related trajectories instead of dedicated excitation trajectories, which is expected to deteriorate the identifiability of the model parameters. For this reason, the presented method uses a sensitivity-based approach allowing model order reduction in the identification process. The resulting model contains only those parameters excited by the excitation trajectory. For process-related trajectories this implies the model being limited to parameters relevant for the process. In experiments with a standard serial-link industrial robot controlled by standard industrial programmable logic control and servo inverters it is shown that the method produces significantly reduced models with a good measure of identifiability and quality.

The present paper deals with an online approach to learn the inverse dynamics of any robot. This is realized by the use of Gaussian Processes drifting parallel along the system data. An extension by a database enables the efficient use of data points from the past. The central component of this work is the implementation of such a method in a controller in order to achieve the actual goal: the feedforward control of an industrial robot by means of machine learning. This is done by splitting the procedure into two threads running parallel so that the prediction is decoupled from the computing-intensive training of the models. Experiments show that the method reduces the tracking errors more clearly than an elaborately identified rigid body model including friction. For a defined trajectory, the squared areas of the tracking errors of all axes are reduced by more than 54% compared to motion without pre-control. In addition, a highly dynamic pick-and-place experiment is used to investigate the possible changes in system dynamics. Compared to an offline trained model, the approximation error of the proposed online approach is smaller for the remaining time of the experiment after an initial phase. Furthermore, this error is smaller throughout the experiment for online learning with parallel drifting Gaussian Processes than when using a single one.

No abstract