Tool: Accessible Automated Reasoning for Human Robot Collaboration Ivan Gavran Ortwin Mailahn Rainer Müller Max Planck Institute for Software Research Group: Assembly Planning Research Group: Assembly Planning Systems Zentrum für Mechatronik und Zentrum für Mechatronik und Kaiserslautern, Germany Automatisierungstechnik gGmbH Automatisierungstechnik gGmbH [email protected] Saarbrücken, Germany Saarbrücken, Germany [email protected] [email protected] Richard Peifer Damien Zufferey Research Group: Assembly Planning Max Planck Institute for Software Zentrum für Mechatronik und Systems Automatisierungstechnik gGmbH Kaiserslautern, Germany Saarbrücken, Germany [email protected] [email protected] Abstract about the underlying reasoners. We present encouraging re- We present an expressive, concise, and extendable domain sults by applying Tool to a case study from the automotive specific language for planning of assembly systems, such and aerospace industry. as industrial human robot cooperation. Increased flexibil- CCS Concepts • Theory of computation → Automated ity requirements in manufacturing processes call for more reasoning; Description logics; • Software and its engineer- automation at the description and planning stages of man- ing → Domain specific languages; Software usability; ufacturing. Procedural models are good candidates to meet this demand as programs offer a high degree of flexibility Keywords assembly planning, automated reasoning, cyber- and are easily composed. physical systems, domain specific language, human-robot Furthermore, we aim to make our programs close to declar- cooperation, industry 4.0, knowledge integration, robotics ative specification and integrate automatic reasoning tools and automation to help the users. The constraints come both from specific ACM Reference Format: programs and preexisting knowledge base from the target do- Ivan Gavran, Ortwin Mailahn, Rainer Müller, Richard Peifer, main. The case of human robot collaboration is interesting as and Damien Zufferey. 2018. Tool: Accessible Automated Reasoning there is a number of constraints and regulations around this for Human Robot Collaboration. In Proceedings of the 2018 ACM domain. Unfortunately, automated reasoners are often too SIGPLAN International Symposium on New Ideas, New Paradigms, unpredictable and cannot be used directly by non-experts. and Reflections on Programming and Software (Onward! ’18), Novem- In this paper, we present our domain specific language ber 7–8, 2018, Boston, MA, USA. ACM, New York, NY, USA, 13 pages. https://doi.org/10.1145/3276954.3276961 “Tool Ontology and Optimization Language” (Tool) and de- scribe how we integrated automated reasoners and planners 1 Introduction in a way that makes them accessible to users which have little programming knowledge, but expertise in manufactur- Software used to be limited to purely logical tasks but it ing domain and no previous experience with or knowledge has become pervasive and now controls many aspects of the systems around us. With the increasing speed at which cyber-physical systems get deployed, software interacts with Permission to make digital or hard copies of all or part of this work for the real world in a more and more autonomous way. personal or classroom use is granted without fee provided that copies The waterfall development model for software was adapted are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights from the more established engineering fields including the for components of this work owned by others than the author(s) must manufacturing industry [Benington 1983], but it didn’t match be honored. Abstracting with credit is permitted. To copy otherwise, or the reality of most software projects. Therefore, iterative and republish, to post on servers or to redistribute to lists, requires prior specific incremental development processes are now the de facto permission and/or a fee. Request permissions from [email protected]. standard in software development. Increasing digitization, Onward! ’18, November 7–8, 2018, Boston, MA, USA automation, and competitive pressure to reduce the time © 2018 Copyright held by the owner/author(s). Publication rights licensed to ACM. to market makes the manufacturing sector move toward to ACM ISBN 978-1-4503-6031-9/18/11...$15.00 iterative and decentralized processes. This is part of a larger https://doi.org/10.1145/3276954.3276961 trend referred to as “Industry 4.0” [McKinsey & Company 44 Onward! ’18, November 7–8, 2018, Boston, MA, USA I. Gavran, O. Mailahn, R. Müller, R. Peifer, and D. Zufferey 2015]. The challenges arising there are very similar to chal- experts, they can be unpredictable for people not familiar lenges of software development. with their underlying working principles. For instance, the In this paper, we look at some of the decision making common understanding of worst case computational com- in human robot collaboration (HRC) in the context of task plexity is not helpful for algorithms based on proof search. assignment for assembly planning. This leads to challeng- Sometimes, having more constraints, i.e. a larger instance, ing questions regarding the choice of the best system setup makes the solver faster (and sometimes not ...). This has led for the lifecycle sequence of the product and commission- to a renewed interest for more predictable systems based on ing related aspects like the safety of humans when working tractable logics. For instance, the Ivy systems [Padon et al. in cooperation with automated systems. Processes that are 2016] uses “only” the Bernays-Schönfinkel class, also known dangerous for humans (when a robot welds, for instance) as EPR, which is NEXPTIME-complete [Lewis 1980]. In this should exclude humans. There are also tasks that particular work, we aim for even more tractable logics. The downside robots cannot execute due to their limitations. Our final goal of using simple logics is that encoding a problem can become is to deliver a correct-by-construction approach to HRC in less intuitive as some constraints need to be reduced to more assembly planning. Instead of relying on humans to know restricted classes. all the regulations and not making any mistake, we rely on Instead of using SMT solvers which are pervasive in the automated reasoning tools integrated in a domain specific programming language and verification community, knowl- language (DSL) to make the critical decisions. We present our edge representation in Tool is based on Owl 2 [Motik et al. first steps in that direction. While the idea of using ontolo- 2012] and the associated reasoners. This choice is motivated gies for assembly planning knowledge isn’t new (see [Raza by the fact that Owl has been chosen by standardization and Harrison 2011], for instance), we realized quickly that authorities [IEEE 2015; Schlenoff et al. 2012] to provide a for- current formal languages are hardly usable for non computer mal basis of robotics. We use the Owl ecosystem to benefit scientists. Developing a DSL became an important step to from these efforts. make the underlying formal model more understandable and Tool is being tested on examples of assembly processes usable. from the aerospace and automotive industries at Zentrum für We have developed the Tool Ontology and Optimization Mechatronik und Automatisierungstechnik gemeinnützige Language (Tool). The goal of Tool is to provide a light- GmbH (ZeMA) within a project aiming to develop more weight interface which helps non-experts to effectively use efficient cyber-physical production technologies. automated reasoning and planning tools. In a sense, Tool is a restriction over existing software, as it only exposes 1.1 A (De)Motivating Example tractable features and emulates some more advanced fea- Among this push for formalization and automation of the tures by paraphrasing them with simpler ones when possi- assembly processes, a key component of our approach are ble. The programmer using Tool first formalizes products, automated reasoners which transform a declarative specifi- processes and resources in Tool’s DSL. Then, Tool checks cation of the problem into a workable solution that meets that suitable resources are provided to perform all processes the specification. The goal of the following experiment was using an automated reasoner. Finally, the resources selected to assess how usable automated reasoners are out-of-the-box by the reasoner are ordered into a schedule. The schedule is for non-experts. optimized according to a linear combination of four objective We asked a fellow at ZeMA to formulate a small prod- functions: cost, duration, probability of stable processes, and uct structure in the Web Ontology Language (Owl 2) with quality of work. The optimization includes time and cost the help of the Protégé editor [Knublauch et al. 2005]. That for scheduled tooling of workers when necessary. While the person had some programming experience, but is not a com- final stage of Tool is a planning problem that subsumes all puter science professional. The given task was modeling the the steps which come before, we still decide to decompose component hierarchy of an aircraft fuselage with 1 skinner, the problem in smaller chunks and solve them separately. 4 formers, 5 stringers and 16 clips. It is a small subset of the The overall complexity remains the same but, from a prac- case study that we will present in Section4. No assembly pro- tical perspective, it makes the