Cargando…

Modular Motion Planner for Robot Arms

The motion planning is one of the main fields of study in robotics since its beginnings. New methods and approaches are constantly published by the scientific community, improving or proposing research routes towards new concepts. Until know teleoperated robots where the main focus of CERN to carry...

Descripción completa

Detalles Bibliográficos
Autor principal: Diaz Rosales, Alejandro
Lenguaje:eng
Publicado: 2019
Materias:
Acceso en línea:http://cds.cern.ch/record/2703282
Descripción
Sumario:The motion planning is one of the main fields of study in robotics since its beginnings. New methods and approaches are constantly published by the scientific community, improving or proposing research routes towards new concepts. Until know teleoperated robots where the main focus of CERN to carry out maintenance work around its particle accelerators and facilities, to reduce the radiation exposition of the workers. New projects in which the level of the robot's autonomy has to increase are appearing. A motion planner is one of the most important elements that would allow the execution of autonomous tasks. This project aims to not only create a functional motion planner for robots but to do it in a way that is easy to update and modified while ensuring a high level of robustness. A modular architecture allows splitting the main problem into several more simple ones. Each of these solutions is done as a module that can also be used outside of the motion planner, for any other application that needs its functionalities. The system can work with any robot arm in which information like the joints constraints, Denavit-Hartenberg parameters or 3D models are known. The motion planner is divided into three main parts. The collision detector that tells if the robot is in contact with the environment or with itself. The path planner that generates a path that takes the robot from a point A to a point B in the space, to generate it the path planner is constantly asking to the collision detector if there is any collision. Finally, the trajectory generator takes the path to transform it into a time scale curve. The collision detection is done with a representation of the objects in voxels and using a GPU (CUDA) to accelerate the detection of collisions between objects. As a first approach to this problem, the path planner can execute several methods, such as RRT*, Informed RRT, BIT*, among others. A comparison of all these methods is done to check and analyze their behavior. The goal of the trajectory generator is to create a motion for each joint of the robot that moves it as fast as possible while ensuring the correct following of the path. Every module is individually validated and in the case of the complete application, with a real robot in a customizable environment.