Optimization-Based Trajectory Planning for Precision Motion Systems and Autonomous Robotic Inspection

Abstract

In industrial environments economic savings are a primary driver for constant quality and productivity improvements. For motion systems in particular, improvements can be achieved with a better choice of motion trajectories. This thesis proposes a set of methods to plan trajectories with applications in precision motion systems and autonomous robotic inspection. First, we propose an optimization-based pre-compensation method to improve the contour tracking performance of precision motion systems. The approach modifies the reference trajectory, while leaving unaltered the built-in low-level controller. The position of the precision motion system is simulated with two data-driven models of different fidelity. A linear low-fidelity model is employed to optimize path traversal time, by manipulating the path velocity and acceleration profiles. Next, a non-linear high-fidelity model is used to refine the previously computed time-optimal solution. Second, we propose an algorithm for the nonlinear iterative learning control problem based on sequential quadratic programming. We iteratively solve quadratic subproblems built by combining approximate nonlinear models and process measurements to find an optimal input for the original system. We demonstrate our method in a trajectory optimization problem for a precision motion system. We present simulations and experimental results to validate the performance of the proposed method, comparing the achieved performance for linear and nonlinear gradient models. We demonstrate experimentally that both methods are capable of simultaneously improving the productivity and the accuracy of a precision motion system. Given the data-based nature of the models, they can easily be adapted to a wide family of precision motion systems. We then address the problem of planning trajectories for autonomous robotic inspection. Volume estimation in large indoor spaces is an important challenge in robotic inspection of industrial facilities. We propose an approach for volume estimation for autonomous systems using visual features for indoor localization and surface reconstruction from 2D-LiDAR measurements. A Gaussian Process-based model incorporates information collected from measurements given statistical prior information about the surface. The volume is measured from the surface model reconstruction. Our algorithm finds feasible motion trajectories for quadcopters which minimize the uncertainty of the volume estimate. We show simulation and experimental results for the surface reconstruction of topographic and industrial data.