Autogeneration and optimization of pick-and-place trajectories in robotic systems: A data-driven approach

Abstract

For manufacturing processes in industries such as aerospace, automotive and electronics, it is essential for robots to perform pick-and-place tasks with high efficiency and accuracy. To this end, we propose a data-driven framework to generate a pick-and-place trajectory that ensures high-accuracy tracking while simultaneously reducing residual vibration, which is particularly valuable for commercial industrial robots with unchangeable control systems. The proposed approach includes both the trajectory generation and the trajectory compensation phases. In the first phase, we plan a pick-and-place trajectory that effectively attenuate residual vibration by minimizing the acceleration energy within a specific frequency spectrum, where the frequency parameters and the time ratio are tuned by Bayesian optimization. In the second phase, we focus on improving the tracking accuracy by incorporating a trajectory compensation term. More precisely, we first learn a Koopman operator-based linear predictor, where a model-agnostic meta-learning framework is introduced to mitigate the demand for massive data from the target system. Then, we calculate the trajectory compensation term using an iterative learning control-based method. The proposed methodology is entirely data driven, enabling its application in various robotic systems and has potential in other manufacturing applications. We demonstrate the approach through high-fidelity simulations on Delta robots – a representative parallel robot, where trajectory generation effectively removes vibrations, and through physical experiments on UR5 robots – a typical serial robot. The results of the experiment show that the positioning accuracy of the three joints of the UR5 robot improved by 94%, 43%, and 96%. • An autogeneration and optimization framework for pick-and-place trajectories. • A trajectory generation method to mitigate residual vibrations. • An ILC-based trajectory compensation method to improve accuracy. • The proposed framework is entirely data-driven.