Intuitive Instruction of Industrial Robots : A Knowledge-Based Approach

Abstract

With more advanced manufacturing technologies, small and medium sized enterprises can compete with low-wage labor by providing customized and high quality products. For small production series, robotic systems can provide a cost-effective solution. However, for robots to be able to perform on par with human workers in manufacturing industries, they must become flexible and autonomous in their task execution and swift and easy to instruct. This will enable small businesses with short production series or highly customized products to use robot coworkers without consulting expert robot programmers. The objective of this thesis is to explore programming solutions that can reduce the programming effort of sensor-controlled robot tasks. The robot motions are expressed using constraints, and multiple of simple constrained motions can be combined into a robot skill. The skill can be stored in a knowledge base together with a semantic description, which enables reuse and reasoning. The main contributions of the thesis are 1) development of ontologies for knowledge about robot devices and skills, 2) a user interface that provides simple programming of dual-arm skills for non-experts and experts, 3) a programming interface for task descriptions in unstructured natural language in a user-specified vocabulary and 4) an implementation where low-level code is generated from the high-level descriptions. The resulting system greatly reduces the number of parameters exposed to the user, is simple to use for non-experts and reduces the programming time for experts by 80%. The representation is described on a semantic level, which means that the same skill can be used on different robot platforms. The research is presented in seven papers, the first describing the knowledge representation and the second the knowledge-based architecture that enables skill sharing between robots. The third paper presents the translation from high-level instructions to low-level code for force-controlled motions. The two following papers evaluate the simplified programming prototype for non-expert and expert users. The last two present how program statements are extracted from unstructured natural language descriptions.

Abstract (Swedish)

With more advanced manufacturing technologies, small and medium sized enterprises can compete with low-wage labor by providing customized and high quality products. For small production series, robotic systems can provide a cost-effective solution. However, for robots to be able to perform on par with human workers in manufacturing industries, they must become flexible and autonomous in their task execution and swift and easy to instruct. is will enable small businesses with short production series or highly customized products to use robot coworkers without consulting expert robot programmers. e objective of this thesis is to explore programming solutions that can reduce the programming effort of sensor-controlled robot tasks. e robot motions are expressed using constraints, and multiple of simple constrained motions can be combined into a robot skill. e skill can be stored in a knowledge base together with a semantic description, which enables reuse and reasoning. e main contribu- tions of the thesis are 1) development of ontologies for knowledge about robot devices and skills, 2) a user interface that provides simple programming of dual-arm skills for non-experts and experts, 3) a programming interface for task descriptions in unstructured natural language in a user-specified vocabulary and 4) an implementation where low-level code is generated from the high-level descriptions. e resulting system greatly reduces the number of parameters exposed to the user, is simple to use for non-experts and reduces the programming time for experts by 80%. e representation is described on a semantic level, which means that the same skill can be used on different robot platforms. e research is presented in seven papers, the first describing the knowledge representation and the second the knowledge-based architecture that enables skill sharing between robots. e third paper presents the translation from high-level instructions to low-level code for force-controlled motions. e two following papers evaluate the simplified programming prototype for non-expert and expert users. e last two present how program statements are extracted from unstructured natural language descriptions.

• author: Stenmark, Maj
• supervisor: Jacek Malec; Elin Anna Topp; Mathias Haage
• opponent: Dr Cakmak, Maya, University of Washington, USA
• organization: Department of Computer Science; Robotics and Semantic Systems
• alternative title: Intuitiv instruktion av Industrirobotar : Ett kunskapsbaserat tillvägagångssätt
• publishing date: 2017-05-02
• type: Thesis
• publication status: published
• subject: Computer Science
• keywords: robot skills, high-level programming, industrial robot, human-robot interaction, natural language, knowledge representation
• pages: 232 pages
• defense location: E:1406, E-huset, John Ericssons väg 4, Lund University, Faculty of Engineering.
• defense date: 2017-05-29 10:15:00
• ISBN: 978-91-7753-297-2
• ISBN: 978-91-7753-298-9
• language: English
• LU publication?: yes
• id: 581e7838-4cf1-45fd-9c4c-d71f099482d9
• date added to LUP: 2017-05-02 12:00:42
• date last changed: 2022-04-11 10:58:27