Path Planning for Industrial Robots Using Evolutionary Algorithms (Masterarbeit)

Abstract

Presently, path planning for industrial robots can be done either online or offline. Online (Teach-In) path planning involves moving the actual robot to the desired positions, and using a GUI or scripting language to define the movements between those positions. This is undesirable for industry use because it occupies the robot for long periods of time, reducing productivity. Offline path planning frees the physical robot, but still requires a large amount of manual labor laying out the desired robot paths. In order to improve the path planning process, previous work has introduced an automatic path planning framework. This framework, based on the KoKo software developed at DLR, offers a robust way to automatically use path planning algorithms in virtual scenarios that accurately replicate the work space setup. This enables the user to quickly adapt a robot to new tasks, or plan paths for many different pickup- and drop points efficiently. Any path planning algorithm can be implemented using this framework.This work aims to develop a path planning algorithm for this framework, in addition to the existing approaches, using an evolutionary approach. While classic path planning algorithms are successful in finding solutions for complex path planning problems, they can consume a lot of computation time because of the very large search space of complex robotic scenarios. This work explores whether an evolutionary algorithm can efficiently find solutions to complex robot pathing scenarios, and whether it can improve upon established path planning algorithms.