Success Stories Success Stories

 

Riding the Crest of Innovation: Model-based Development of a Collaborative Transport Robot Fleet

As a research project funded by the Federal Ministry for Education and Research (BMBF), Collaborative Embedded Systems (CrESt) focuses on the development of complex embedded systems that have to cooperate efficiently to fulfil given tasks under varying contexts and with different constituents. The project’s goal is to define methods and architectures using model-based systems and system context descriptions for dynamic and scalable applications. Collaborative embedded systems (CES) and adaptive system architectures will have a significant impact on technological development in the near future. Through these systems and their architectures, current factory processes can be redefined and profit from a high degree of automation. They can also assist factories in reacting to changing production factors in a more flexible way. For instance, learning machines may take over repetitive tasks previously done by humans to optimize these processes in a goal-oriented manner. Additionally, they may take control of the work of central order management systems in order to coordinate and organize their transport orders autonomously. Centered on model-based system designs, analyses can be run to record all features a system will need with the emphasis on safe collaboration. The cases where such technology would be useful include autonomous robots, learning control systems, and adaptable factories amongst many others.

Model-Based Development for Collaborating Embedded Systems

proANT 436 at the technology firma, DYCONEX
Figure 1: proANT 436 at the technology firma, DYCONEX

InSystems Automation GmbH and Model Engineering Solutions GmbH are two of 22 international associates from industrial and scientific backgrounds who are working together in the joint research project in order to open up access to new technology and methods to develop CES as well as collaborative system groups (CSGs) such as a fleet of autonomous robots. Collaborating systems are defined by working together efficiently as opposed to competing amongst each other. CrESt is divided in six co-called “Engineering Challenges” and six interdisciplinary topics. The first three engineering challenges are concerned with the overarching question of how to best design the architecture of flexible, dynamic, and adaptive systems. In the project aspect of dynamic architectures, it is, among others, the task of InSystems Automation and Model Engineering Solutions to investigate options to integrate a robot into a pre-existing fleet using mechanisms such as Plug & Play without the need to pause the whole production process. In addition, InSystems’ goal is to improve their so-called proAnt transport robots, which have been in production since 2012. The general focus lies on improvements in technology and creating and evaluating new approaches to their overall concept. More precisely, key topics include decentralized fleet management, communication among individual systems, and analysis of environment data.

Due to the number of difficulties that result from the high complexity of the involved scenarios that have to be incorporated, the development of adaptive systems requires a well-grounded approach. A fleet of robots has to react to dynamic changes in the policy of the manufacturing execution system or the number and nature of its members in such a way that the overall functionality and efficiency of the CSG is safeguarded. The consistent application of a model-based development process for automation systems offers a variety of beneficial properties. Most importantly, the specification of the CSG and collaborative AGV controllers (CACs) in the form of executable models allows for a fully virtual representation (digital twin) of the robot fleet members and their collaborative adaptive behavior. This virtual representation provides a sound foundation to efficiently develop, maintain, and extend the actual system along with its hardware, software, and mechanical components. To exploit its full potential, a model-based approach first and foremost relies on the reusability of models and test beds over the different developments phases. These include function, system, and system component development. Secondly, the model-based development process builds on a fully integrated toolchain that highly automatizes the associated development activities including requirements management, modelling, and simulation as well as integrated quality assurance tasks, most notably, model-based static analysis and requirements-based testing tools.

Adaptive System Architectures for a Fleet of Transport Robots

In the design and maintenance of adaptive embedded systems, the consideration of the system’s context is a key aspect of nearly all modeling and analysis methods. In the context of the use case of collaborating transport robots, the job assignment procedure is the most critical part of collaboration. For this reason, defining and applying one or a set of strategies is necessary to determine the responsible robot for the job execution when a job arises. In other words, the CSG must be able to fulfill all the jobs requested by the factory, while adapting its behavior to the contextual changes. In this case, the question arises of how a CSG can distribute the jobs among themselves in a way that they fulfill both local goals, which are goals that are specific to individual robots such as a minimal battery state of charge, as well as global goals, which are goals determined by a suitable production policy given by the manufacturing execution system. The production policy for collaborating transport robots specifies a common goal for the CSG, which is broadcast to the fleet by a manufacturing execution system.

Fleet of proANT AGVs at Bierbaum Unternehmensgruppe, transporting open barrels
Figure 2: Fleet of proANT AGVs at Bierbaum Unternehmensgruppe,
transporting open barrels

In cooperation with InSystems, software engineers at Model Engineering Solutions studied and evaluated four specific global goals that are dynamically broadcast by the manufacturing execution system, namely economy (to minimize the total distance driven by all CACs), robustness (to keep the job queue lengths of each robot as small as possible), performance (to maximize the number of jobs executed per time unit), and maintenance (to distribute the tasks such that all robots drive a similar distance). These global goals are encoded using a suitable bidding parameter vector. It is important to note that these goals may change at run time without any prior information at design time. The dynamically varying goals have to be realized as concrete CSG strategies, which are autonomously resolved by the robot fleet. Other global CSG goals, including timing restrictions on the job distribution as well as local goals, also need to be taken into account. Any dynamic change in the context may trigger a reconfiguration phase. This is the case if a new job is broadcasted by the manufacturing execution system, if a robot joins or leaves the fleet, or if a new obstacle is detected. Under these circumstances, the CSG needs to reconfigure itself in order to still be compliant with the CSG’s goals. For instance, depending on the given strategy, the job queues handled by each CAC have to be adapted or even redistributed, whenever a robot leaves the fleet without having completed its task list. Tasks of this kind are realized by suitable reconfiguration units.

Modelling of a Transport Robot Fleet in Simulink

proAnt Robot
Figure 3: proAnt Robot

InSystems and Model Engineering Solutions jointly developed a MATLAB/Simulink model of an adaptive fleet of collaborating transport robots such as the proANT robots developed by InSystems. This model was designed to capture the desired adaptive system behavior and to deal with the abovementioned CSG goals and challenges more effectively. The domain-independent semi-formal language Simulink is suitable to describe the CSG/CAC behavior of the robot fleet and its members as well as the context including the manufacturing execution system. Moreover, Simulink models are able to interface with typical robot middleware or communication frameworks such as the Robot Operating System (ROS).

Simulink provides a platform to design, simulate, and validate the behavior of a (dynamical) system on various abstraction levels including function/specification, system, and software models. The tool is widely used in industry as it provides a domain-independent modeling tool for dynamic systems. Typical applications include signal processing, control engineering problems, and systems engineering. In particular, Simulink provides several simulation modes ranging from quasi-continuous to discrete or event-based execution and sampling rates, various solver options, which allow for tailored trade-off between precision, memory consumption, and execution time, and it supports common data type concepts including floating point, fixed point, and enumeration types. Together with its built-in library and various add-ons, including finite state machines, domain-specific model integration, and FMI support, Simulink is well-suited as a tool to quickly create, simulate, and test prototypes of CSGs. In addition, its high reuse potential across various development stages is highly beneficial.

During a model-based development process, it is also essential to follow modeling guidelines in order to ensure functionality, maintainability as well as an overall efficient workflow. For instance, the system decomposition model must only contain subsystem blocks and signal routing elements. In particular, no numerical computations are performed in a decomposition model. This ensures that the CSG requirements can be completely mapped to requirements on each robot’s CAC components. Moreover, decomposition models must specify coherent components with bounded complexity, for which suitable measures are available. Modeling guidelines further address many aspects including safety topics, variant control and strong data typing. During the course of this project, compliance was checked and corrected automatically using suitable static analysis tools like the MES Model Examiner.

Example of a Simulink CAC model with main components such as reconfiguration units and task distribution logic
Figure 4: Example of a Simulink CAC model with main components such as reconfiguration units and task distribution logic

From Formalized Requirements to Automatic Assessments using MTest/MARS

Requirements-based testing with formalized specifications such as MARS
Figure 5: Requirements-based testing with
formalized specifications such as MARS

In order to develop a practical method to implement, operationalize, and validate the adaptive behavior of the CSG, the focus was placed on the Simulink-based CSG prototypes. The above mentioned system requirements must be validated using a test-driven approach and monitored at run time to indicate possible system failures in order to conduct suitable countermeasures. Policy changes induce protocol changes, which need to be properly addressed in the system specification. Most importantly, the expected adaptive system response subject to dynamically varying policies of the manufacturing execution system must be fully captured in the CSG requirements. Consistent system specifications must be well-defined, unified, and uniquely comprehensible, which directly leads to the need for formalized requirements.

Compared to natural language-based approaches, which are still widely used in practice, formalized requirement formats give rise to unambiguous representations of CSG requirements. Moreover, formalized requirement formats such as the MTest Assessable Requirements Syntax (MARS) can be fully integrated with the model-based approach in the sense that state- or event-based triggers and the required signal response can be fully defined using references to model entities such as signal specifications or design parameters. In conjunction with the efficient definition of appropriate test cases, virtual validation of adaptive CSG behavior can be automatized based on automatic test execution and assessment. In particular, MARS bridges the gap between requirements based on formal languages and requirements that can be easily formulated and accessed by system engineers of different backgrounds using natural language concepts. As part of the model testing tool MES Test Manager (MTest), MARS further provides a basis for various aspects of test automation in the context of model-based development, including automatized generation of assessments, test cases, and monitors.

Overall, the development and maintenance of a CSG as described above can be based on a fully virtual counterpart in the form of interacting Simulink models that represent the behavior of the fleet of transport robots. As a result, quality assurance methods, for example requirements-based testing methods, can be based on a fully virtual prototype of the robot fleet using tools such as MTest, which highly automatize quality assurance tasks. This so-called “frontloading” approach assists in finding design and implementation errors very early in the development process. Furthermore, it facilitates the rapid testing and validation of the integration of new robot types or the implementation of new collaboration protocols.

About InSystems Automation:

InSystems Automation develops innovative automatic solutions and special machines for production, material flow and quality tests. Our range of services includes all tasks, from creating the specification sheet, to electrical projects, installation and programming and commissioning, maintenance and service. Our customers have a single and professional contact person, from planning to the completion of machines and plants. The company was founded in 1999 by the managing directors Henry Stubert and Torsten Gast and has grown constantly since then. Currently, more than 50 employees work at InSystems. The company is located in the science center Berlin-Adlershof and has offices, a workshop, an online shop and a showroom for industry 4.0. Since 2012, InSystems specializes in the production of autonomous navigating transport robots, which are designed for loads from 30 to 1.000 kg, according to customer request, and implemented as a fleet into an existing production control. The transport robots are developed under the name proANT.

About MES: Software Quality. In Control.

Founded in 2006 in Berlin (Germany), software company Model Engineering Solutions GmbH (MES) offers solutions for integrated quality assurance of software projects. MES supports its customers in developing model-based software in compliance with industry standards such as IEC 61508, ISO 25119, and ASPICE. The MES Tool Chain comprises four complementary tools for all phases of the model-based software development process: the MES Quality Tools. The MES Model Examiner® (MXAM) conducts automated checks to verify compliance with modeling guidelines for MATLAB Simulink®/Stateflow®, Embedded Coder®, TargetLink®, and ASCET® models. The MES Test Manager® (MTest) efficiently implements requirements-based unit testing in model-based development. MES M-XRAY®’s fast and precise structure and complexity analysis gives you complete transparency of your Simulink®, Stateflow®, Embedded Coder®, and TargetLink® models. MES Quality Commander® (MQC) evaluates the quality and product-readiness of your software and delivers key decision-making data throughout the product development lifecycle. MES Model & Refactor® (MoRe) supports users in modeling with MATLAB Simulink® by simplifying and accelerating model editing and reducing monotonous work steps. The MES Test Center team supports its customers with testing services from requirements management through setting up test specifications and automated test evaluation to quality monitoring. The MES Academy provides training classes and company-specific consulting services and projects. The consultants offer support in introducing and improving model-based development processes to fulfill standards such as IEC 61508, ISO 26262, and ASPICE. MES’ clients in the automotive industry include 16 of the 20 largest OEMs worldwide and their suppliers. In addition to this, the number of MES’ customers in the automation technology field continuously grows. MES is a dSPACE Strategic Partner and a MathWorks and ETAS Product Partner. The MES Academy collaborates with SAE International.