Context Aware Sliding Work Sharing for Human-AI Collaboration in Logistics Domain

The final publication is available at Springer: https://doi.org/10.1007/978-3-032-05681-8_24 2 Context Aware Sliding Work Sharing for Human-AI Collaboration in Logistics Domain Sebastian Scholze1, Ana Correia1, Gunnar Große Hovest1 1 ATB Institut für angewandte Systemtechnik Bremen GmbH, Wiener Str. 1, 28359 Bremen, Germany {scholze,correia,gr-hovest}@atb-bremen.de Abstract. This paper introduces a novel framework for context-aware sliding work sharing (SWS) to enhance human-AI collaboration. It addresses the challenge of dynamically balancing human and machine contributions by leveraging real-time contextual information and machine confidence levels. The proposed system integrates advanced context awareness mechanisms — comprising context modelling, monitoring, and extraction — with adaptive SWS management to facilitate informed task delegation between humans and AI/robots. By adjusting autonomy levels based on situational complexity and user expertise, the approach aims to improve decision-making, efficiency, and trust in AI-supported environments. A logistics-based use case demonstrates the practical applicability and potential benefits of the system. Keywords: Context awareness · Sliding work sharing · Human-AI collaboration · Logistics · Adaptive autonomy · Decision support 1 Introduction AI and robotics are likely to be the most powerful means for radical improvement of working conditions in diverse domains such as industry, logistics, healthcare, construction, agriculture, education etc. AI and robots can support human operators in diverse tasks starting from difficult and tedious manual labour tasks up to complex decision-making tasks. This paper addresses the approach of the AI4Work project [1] to support the operation of complex systems with general questions such as how to increase efficiency of work, reduce the stress upon employees or increase confidence in decision-making tasks. In particular, the main question that we aim to answer is how to optimally share the work between humans and AI/robots. We provide an overview of the topics relevant for this work, namely on context awareness and sliding work sharing (section 2), follow with the vision for the proposed Context Aware Sliding Work Sharing (section 3) and support this with the Context Awareness and Sliding Work Sharing tools (section 4). Finally, we describe a potential application in the logistic sector and some conclusions. 3 2 Overview of State of the Art 2.1 Context Awareness The context under which work activities are carried out is of a high relevance for both (a) accuracy/reliability of AI recommendations/activities and (b) acceptance of these actions by the users. (a) The context under which data are acquired and processed may have a high influence upon collaborative decision making. As indicated in [2], the meaning of many concepts heavily depends on some implicit context, and changes in that context can cause radical changes in concepts. Incremental concept learning in such domains requires the ability to recognize and adapt to such changes. This is particularly the case in modern organisations, where the context under which the work processes are carried out changes very often. For example, faults in manufacturing and logistics processes have different meaning and weight if the equipment is in installation and ramp-up, or in normal (stationary) production phase, or if there is a change in a configuration of a production line due to changes in demands, etc. (b) The interpretation of the context under which a recommendation by AI is proposed or actions by AI/robots are performed is important to better understand why a specific recommendation/actions is made by the AI/robot module, i.e. to allow for an improved traceability of the AI recommendations. It is expected that this would lead to better acceptance of AIsupported decision making and activities in organisations. Context Awareness is a concept propagated in various domains, such as ubiquitous computing and AI. It is the idea that computers can be both sensitive and reactive, based on their environment [3]. It is difficult to find a single definition for the notion of context, but its importance in communication, categorization, intelligent information retrieval and knowledge representation has been recognized for many years. In the AI domain, the concept of context is usually defined as the generalization of a collection of assumptions [4], [5]. A common, pragmatic definition for context-aware applications, defines context as “any information that can be used to characterize the situation of an entity. An entity is a person, place, or object that is considered relevant to the interaction between a user and an application, including the user and application themselves” [6], [7]. The current research on knowledge context is primarily oriented towards capturing and utilization of contextual data for actionable knowledge [2], [8]. A number of systems to handle context awareness were proposed by the research community [9], [10], [11]. Thereby an important aspect is to make the data available for contextual analysis and processing. Various solutions for monitoring and ingesting data into contextual analytics services are available, e.g. U-QASAR [12], SAFIRE Context Monitoring Framework [13] or FIware Context Broker [14]. The key elements of context awareness solutions are context models (that describe the situation) and context extractors which identify often in real time the current context to which the ICT environment has to adapt. The basis for context-aware applications is a well-designed Context Model. As context integrates different knowledge sources and binds knowledge to the user to guarantee that the understanding is consistent, context modelling is extensively investigated within knowledge management research [3]. 4 Different kinds of contexts should be represented in a common “language” when possible, but the representation must be extensible enough to support domain and application specific concepts. Typical context modelling techniques include key-value models, object-oriented models, and ontological methods [15]. A semantic model (or ontology) provides a representation flexible enough to support common modelling of context in a structured way, as well as domain specific extension to the model, thus it is chosen for representation purposes. Ontology based modelling is considered the most promising approach, as it enables a formal analysis of the domain knowledge, promoting contextual knowledge sharing and reuse in a ubiquitous computing system, and context reasoning based on semantic web technologies [16] [17]. Up to now there were only limited “industrial driven” attempts to provide harmonised modelling of context under which data from human-AI/robot collaboration are generated. The problem to be solved is how to extract context from human-AI/robot collaboration. Since it is planned to model the context with ontologies, the context extraction is mainly an issue of context reasoning and context provisioning: how to inference high level context information from low level raw context data [18] and [19]. Based on the formal description of context information, context can be processed with contextual reasoning mechanisms [20] [21]. There are three main categories of reasoning, deductive reasoning, event-condition-action reasoning and statistical reasoning which can be distinguished. Context awareness is of special importance in the human-AI/robot collaboration in various domains and used under dynamically changing conditions and by various users [22]. The most challenging aspect of the application of the context awareness for human-AI/robot collaboration is an effective acquisition/collection of data needed to extract the current context. Therefore, the advanced collection of data, is a basic prerequisite for an effective application of this approach in different sectors. The key challenge is to identify cost-effective ways to obtain data needed for context extraction. The context awareness features of the AI4Work services consider as basis important outcomes of solutions developed in the projects SelfLearning [19, 22], K-NET, SAFIRE [23], DIVERSITY [24], and in SmartCLIDE [25]. The context extractor will be re-used, but the context models will have to be adapted and extended to meet process requirements (e.g., to extend the model, in addition to concepts such as location, time, activity, with concepts relevant for processes, all other concepts that can be relevant for a human operator, etc.). 2.2 Sliding Work Sharing (SWS) The AI4Work project investigates practical methods and tools for optimal sharing of work between humans and AI/robots. In this context, "sliding work sharing" is defined as an approach where the balance between human and machine activities varies during the operation, depending on the situational context, machine-based confidence levels and human interactions [1]. This term and definition were inspired by the similar definition of "sliding decision making", as presented in [27]. The term "sliding work sharing" (SWS) was introduced by the AI4Work project proposal [1], therefore no prior scientific literature can be found about SWS. However, the related concept of "sliding autonomy" could be identified, which is considered 5 highly relevant as basis for AI4Work's research on SWS. Sliding autonomy, also called "adjustable autonomy", was researched extensively in the domains of robotics and (multi-)agent systems [27]. Instead of having fixed modes like "robot working autonomously" or "human teleoperating the robot", the aim is to vary the level of autonomy during operation, depending on the respective situation. The motivation for this approach is that, on the one hand, highly autonomous robots can help reducing the mental load of human operators [28] and they can also be used in hostile environments, where humans may not be able to work [29] [30], but on the other hand, it is hardly possible (or very time-consuming) to program robots in a way that they can react to all potential challenges that may happen in dynamic real-world environments [29], [28], [31]. Therefore, humans should at least monitor the robots' work and should be able to intervene in case of robots "being stuck" or in case of potential safety hazards. Another motivation for sliding autonomy may be to dynamically adjust the robots' autonomy level depending on the skill level of the human who is controlling the robot [32]. Different scenarios regarding sliding autonomy are discussed in the existing literature, involving either single robots (e.g. [32]) or teams of robots, either being supervised by humans (e.g. [29] [28]) or working with humans in peer-to-peer mode (i.e. humans are working together with robots on the same task or can even get tasks assigned by robots, e.g. [27] [31]), solving different experimental tasks. Aiming to simplify the description, the example of “one human and one robot working together” is used in the following, however the general principles apply also to multi-robot/multi-human teams. A variety of potential levels of autonomy may be differentiated, e.g. considering the ten levels of automation suggested in [33]. However, when considering sliding autonomy of robots, the three main levels that are most relevant can be summarized as follows: • Fully autonomous: the robot is working autonomously, without any human involvement. • Intermediate: the robot is in general working autonomously, but a human can temporarily provide support or take over control. • Manual operation: the human operates the robot, having full control over it. A temporal human involvement, as indicated in the intermediate autonomy level above, is typically caused by an unexpected issue that either prevents the robot from continuing its work or that decreases the performance of its work. Such unplanned human involvements may be "triggered" by either human or robot, so two basic scenarios can be differentiated: • The robot itself detects an issue and informs the human that it requires support. This may happen for different reasons, e.g. o the robot is blocked/trapped or is lost in the environment [27], o considering a given task, the robot has a low level of "self-confidence" regarding its ability to succeed [30], o a timeout occurred, i.e. the robot did not succeed after a specific time or number of tries [29], [31], o the robot actively assigns a task to a human, which is relevant in peer-to-peer human-robot teams [27]. • The human pro-actively intervenes because they observe one of the following: 6 o the robot's work seems inefficient and thus efficiency improvements are possible through human intervention [29], o the robot is behaving incorrectly [31], o the robot may become a safety hazard (e.g. due to potential collisions with surrounding objects, other robots, or even humans) [29], [30], [32], o there may be a risk of losing the robot, which is especially relevant when working in hostile environments where humans can only teleoperate the robot [30]. Different studies about sliding autonomy did experiments (either physical or simulated) to compare the performance (including required time, error rate, workload for human operator) of sliding autonomy compared to robots being either fully autonomous or under full control of humans, e.g. [27-29, 31, 32]. Although these experiments usually comprise only a limited number of repetitions of a very specific task, some general patterns can be observed. Sliding autonomy typically leads to a higher success rate compared to fully autonomous robots (as humans can help in case robots fail), as well as to faster task solving compared to fully manual operation (because it does not require the human to control every step in detail, which can be tedious and is especially difficult when controlling a multi-robot team). The downside may be, on the one hand, slower task solving compared to fully autonomous robots, and on the other hand, higher failure rates compared to the robot being under full control of a human. Sliding autonomy can thus be considered a compromise/trade-off approach, aiming to find the ideal balance between human and machine activities depending on the respective situation, as it is the goal of SWS. One strategy may be to increase the autonomy level step by step, thus increasing the efficiency and reducing the human workload, but only until the performance of the robot starts to deteriorate [30]. Another aspect to be considered is that the ideal degree of autonomy may be dependent on the experience/skills of the human who is controlling and supporting the robot [28]. When trying to transfer sliding autonomy approaches from the robotics domain to human/AI work sharing scenarios, a basic difference must be considered: a human can rather easily see that a physical robot requires help or could become a safety hazard, but this may not be as easy or even impossible for a human working together with (and observing) an AI. To tackle this challenge, the AI4Work project aims to provide realtime SWS support by continuously (1) monitoring the current context of the work environment, (2) assessing if the AI was properly trained for this kind of situation and (3) suggesting (based on the assessment and additional context information) to which extent the AI may work autonomously and to which extent the human should be involved. 3 Sliding Work Sharing Vision The vision of the AI4Work project is to improve communication and collaboration between humans, AI and robots. AI4Work intends to investigate practical methods and tools for optimal sharing of work between humans and AI/robots (the project will specifically focus upon AI solutions for diverse tasks including AI for intelligent 7 robotic systems). Due to the high level of uncertainty in modern organisations (in industry but also in healthcare, education, agriculture [34], [35] 1 ) an appropriate balance between human and machine activities (e.g., decision making [36] [37] must be found. The key assumption is that to cope with the required flexibility and dynamics, SWS, is likely to be the most appropriate for modern organisations. One of the key challenges in AI-supported working is an “appropriate interpretation of a context guiding machine or human to better understand the proposed support/recommendations/decisions” [38]. Due to the required high dynamics and flexibility in modern organisations, the context, under which work is carried out, changes rapidly. It is, therefore, important to identify online the current context and adapt AI/robots support based on it, and, at the same time, help operators understand the action taken or decision proposed by AI/robots. It is expected that this would lead to changes in attitudes towards AIand robot-supported work in organisations [39]. 4 Tools to Support Context Aware Sliding Work Sharing 4.1 Context Awareness The Context Awareness component is a core element in rendering AI and robotic systems context-sensitive. It facilitates the extraction, interpretation, and representation of the operational circumstances under which collaborative work occurs, thereby enabling adaptive actions and recommendations aligned with the current environment. This component comprises three key modules: • Context Model: Serves as a foundational knowledge base for representing collaborative human-AI/robot work across diverse scenarios. It defines a structured set of concepts and relationships that describe the stages, entities, attributes, and stakeholders involved in such collaboration. • Context Monitoring: Continuously acquires raw data from sensors, systems, and various knowledge sources to detect changes in the environment. It identifies dynamic factors that influence the collaborative processes. • Context Extraction: Processes the monitored data in alignment with the Context Model to derive the current context. This enables comparisons with historical contexts and supports informed decision-making, optimization, and system reconfiguration. Within the AI4Work project, the Context Model is employed as a baseline to capture and formalize knowledge about human-AI/robot collaboration. The integration of Context Monitoring and Extraction services allows the system to identify and interpret context changes, ensuring adaptive and context-aware behaviour. The proposed method establishes mechanisms for: 1 According to [35] “When humans depend on automation to get their work done, they must be able to anticipate what happens, because they, not the machines, are responsible. ” and “Computer scientists should build devices to enhance and empower - not replace - humans.” 8 • Monitoring contextual parameters and detecting changes relevant to collaborative scenarios; • Analysing quantitative relationships that affect context interpretation from multiple perspectives (e.g., user-centric, service-centric views); • Defining entities and parameters critical for context monitoring; • Investigating how monitoring mechanisms vary depending on the perspective adopted, as delineated by the Context Model. By coordinating continuous monitoring and context extraction, the system transforms raw data into meaningful contextual knowledge. This capability is crucial for enabling adaptive decision-making and optimizing collaborative processes in dynamic humanAI/robot environments. Key Challenges. The key challenges in Context Monitoring and Extraction components are: • Can we retrieve all information from all needed sources? AI4Work approach: the Context Monitoring solution defines a generic interface with an extendable and configurable standardised process. The question is can we really retrieve all information from the environment and AI4Work environment needed by the defined context model? • Can we extract the context needed by the other AI4Work modules, based on the monitored information? AI4Work approach: Using the context model, the monitored data is evaluated and the context extracted. Context Model Concept. The AI4Work project employs an ontology-based approach for context modelling, leveraging ontologies’ flexibility, expressiveness, and extensibility. Ontologies ensure a shared semantic understanding of context data across systems, enabling reasoning mechanisms to infer additional knowledge from implicitly stated information. Fig. 1 presents this approach. Fig. 1. Context Awareness Approach Inner Loop Outer Loop Context Extraction Context Monitoring Context Provision Context Modelling Context Model 9 To identify the current contexts of services and processes, as well as the current context of the user in some situations, the proposed solution uses as information sources: • data from the AI4Work components • data from existing systems and sensors • available knowledge from different systems This information is used to extend base/core ontologies to meet the contexts of the systems in question and model the entities, and relationships between them. Context Monitoring Concept. The objective of the Context Monitoring service is to transform raw sensor and system data into aggregated, structured monitoring data suitable for further processing. To achieve this, the service enables the monitoring of enterprise systems via various interfaces. The core of the Context Monitoring service is a modular monitoring process that follows a standardized, extendable, and configurable architecture (Fig. 2). This process comprises three primary modules [19]: • Monitoring Module: Responsible for acquiring data from systems and devices within enterprise environments via the Data Access Layer. Distributed monitoring services interact with this module to relay collected information. The monitoring services are designed to be extendable and configurable for diverse systems, without requiring tight coupling with other modules. • Parser Module: Contains content parsers tailored to the diverse data formats and types captured by the monitoring services. The parsers enable access to heterogeneous data sources and may extract environmental properties relevant to context understanding. This module serves as the data interface for subsequent analysis. • Analyser / Monitoring Data Builder Module: Correlates the parsed monitoring data, potentially incorporating environmental parameters, and constructs standardized monitoring datasets. These datasets are stored and provided to the Context Extraction service or other components requiring monitoring insights. Fig. 2. Context Monitoring Process Existing Device-centric Infrastructure Data Access Layer Monitoring Parser Analyser Monitoring Repository Monitoring Data 16 TERM LOW := (0, 1) (5, 1) (9, 0); TERM MEDIUM := (5, 0) (9, 1) (11, 1) (15, 0); TERM HIGH := (11, 0) (15, 1) (20, 1); Rather than having static values, we have (partly overlapping) intervals and an input value may partly belong to two categories at the same time. Below is an example of the fuzzy rules above mentioned: IF noOfTrucksInQueue IS LOW THEN suggestedWorkSharingApproach IS AI_AUTONOMOUSLY; IF noOfTrucksInQueue IS MEDIUM AND positionOfTruckToBePrioritized IS NEAR_THE_FRONT_OF_THE_QUEUE THEN suggestedWorkSharingApproach IS HUMAN_ON_THE_LOOP; IF noOfTrucksInQueue IS HIGH AND positionOfTruckToBePrioritized IS IN_THE_BACK_OF_THE_QUEUE THEN suggestedWorkSharingApproach IS HUMAN_MANUALLY; After receiving the “sliding decision input data” from the request, the SWS parametrizes and fuzzifies these inputs based on the membership functions specified. It then evaluates the rules using a fuzzy inference system and it calculates defuzzified output (crisp decision results). These results are then extracted and made available for use by other components. 6 Conclusions This paper presented the concept and initial implementation of a context-aware Sliding Work Sharing (SWS) framework designed to support effective human-AI collaboration, illustrated by an application scenario from the logistics sector. By integrating dynamic context awareness with adaptive autonomy mechanisms, the AI4Work approach enables flexible redistribution of tasks between humans and AI/robot systems based on situational context, confidence levels, and user expertise. The framework’s modular design — featuring ontology-based context modelling, realtime monitoring and extraction, and fuzzy logic-based SWS decision making — demonstrates strong potential to improve operational efficiency, reduce cognitive load on human workers, and increase trust in AI-driven systems. The demonstrated yard management scenario highlights the feasibility and value of this approach in real-world logistics operations. Future research directions will focus on several key areas: • Scalability and Generalization: Extending the context-aware SWS framework to support more complex and larger-scale industrial scenarios, including multi-agent systems and cross-domain applications (e.g., healthcare, manufacturing, agriculture). • Context Model Refinement: Enhancing the ontology to support more nuanced contextual features, such as emotional state, team dynamics, or organizational constraints, while also improving the cost-efficiency of data acquisition. 17 • Human-Centric Evaluation: Conducting longitudinal studies to better understand human acceptance, cognitive load, and trust dynamics in AI-assisted work environments under varying autonomy levels. • Learning and Adaptation: Incorporating machine learning techniques to enable continuous learning from user behaviour and feedback, thereby improving the system’s ability to predict optimal work sharing configurations. • Ethical and Legal Considerations: Exploring regulatory, ethical, and transparency aspects related to dynamic work sharing between humans and AI, particularly in high-stakes decision-making contexts. By addressing these directions, the AI4Work approach aims to contribute significantly to the development of trustworthy, flexible, and human-centred AI solutions for the workplace of the future. Acknowledgements. The research described has been carried out as part of the AI4Work Project, which has received funding from the European Union’s Horizon Europe Research and Innovation Programme under grant agreement No. 101135990. References [1] ‘AI4Work - Human-centric Digital Twin Approaches to Trustworthy AI and Robotics for Improved Working Conditions - Grant agreement ID: 101135990’, CORDIS | European Commission. Accessed: Jul. 05, 2024. [Online]. Available: https://cordis.europa.eu/project/id/101135990 [2] H. J. Ahn, H. J. Lee, K. Cho, and S. J. Park, ‘Utilizing knowledge context in virtual collaborative work’, Decision Support Systems, vol. 39, no. 4, pp. 563–582, Jun. 2005, doi: 10.1016/j.dss.2004.03.005. [3] S. Scholze, O. Kotte, D. Stokic, and C. Grama, ‘Context-sensitive decision support for improved sustainability of product lifecycle’, presented at the Intelligent Decision Technologies 2013, 2013. doi: 10.3233/978-1-61499-264-6-140. [4] P. Brézillon, ‘Context in problem solving: a survey’, The Knowledge Engineering Review, vol. 14, no. 1, pp. 47–80, May 1999, doi: 10.1017/S0269888999141018. [5] L. Serafini and P. Bouquet, ‘Comparing formal theories of context in AI’, Artificial Intelligence, vol. 155, no. 1, pp. 41–67, May 2004, doi: 10.1016/j.artint.2003.11.001. [6] Anind K. Dey, ‘Providing architectural support for building context-aware applications’. Georgia Institute of Technology, Jan. 01, 2000. [7] A. K. Dey, Abowd ,Gregory D., and D. Salber, ‘A Conceptual Framework and a Toolkit for Supporting the Rapid Prototyping of Context-Aware Applications’, Human–Computer Interaction, vol. 16, no. 2–4, pp. 97–166, Dec. 2001, doi: 10.1207/S15327051HCI16234_02. [8] T. Yang, T.-Y. Xiao, and L. Zhang, ‘Context-centered design knowledge management’, vol. 10, pp. 1541–1545, Dec. 2004. [9] P. Bellavista, A. Corradi, R. Montanari, and A. Toninelli, ‘Context-aware semantic discovery for next generation mobile systems’, IEEE Commun. Mag., vol. 44, no. 9, pp. 62–71, Sep. 2006, doi: 10.1109/MCOM.2006.1705981. [10] S. Kim, E. Suh, and K. Yoo, ‘A study of context inference for Web-based information systems’, Electronic Commerce Research and Applications, vol. 6, no. 2, pp. 146–158, Jun. 2007, doi: 10.1016/j.elerap.2006.11.003. [11] J.-W. Chang and Y.-K. Kim, ‘Design and Implementation of Middleware and Context Server for Context-Awareness’, in High Performance Computing and Communications, 18 M. Gerndt and D. Kranzlmüller, Eds., Berlin, Heidelberg: Springer, 2006, pp. 487–494. doi: 10.1007/11847366_50. [12] O. Kotte, A. Elorriaga, D. Stokic, and S. Scholze, ‘Context Sensitive Solution for Collaborative Decision Making on Quality Assurance in Software Development Processes’, presented at the Intelligent Decision Technologies 2013, 2013. doi: 10.3233/978-1-61499-264-6-130. [13] S. Scholze, A. T. Correia, and K. Nagorny, ‘Services for development of Situational Aware Intelligent PSS’, presented at the 2017 International Conference on Engineering, Technology and Innovation (ICE/ITMC), IEEE, 2017, pp. 1154–1161. [14] FIWARE, ‘Core Context Management’, FIWARE Catalogue on GitHub. Accessed: Jul. 19, 2024. [Online]. Available: https://github.com/FIWARE/catalogue/tree/master/core [15] T. Strang and C. Linnhoff-Popien, A Context Modeling Survey. 2004. [16] R. Glassey et al., Towards a Middleware for Generalised Context Management. 2003, p. 52. [17] F. Paganelli, G. Bianchi, and D. Giuli, ‘A Context Model for Context-Aware System Design Towards the Ambient Intelligence Vision: Experiences in the eTourism Domain’, in Universal Access in Ambient Intelligence Environments, C. Stephanidis and M. Pieper, Eds., Berlin, Heidelberg: Springer, 2007, pp. 173–191. doi: 10.1007/978-3-540-710257_12. [18] S. Scholze, J. Barata, and D. Stokic, ‘Holistic Context-Sensitivity for Run-Time Optimization of Flexible Manufacturing Systems’, Sensors (Basel), vol. 17, no. 3, p. 455, Feb. 2017, doi: 10.3390/s17030455. [19] D. Stokic, S. Scholze, and O. Kotte, ‘Generic Self-Learning Context Sensitive Solution for Adaptive Manufacturing and Decision Making Systems’, presented at the ICONS14 - The Ninth International Conference on Systems, 2014. [20] J. Forstadius, O. Lassila, and T. Seppanen, ‘RDF-based model for context-aware reasoning in rich service environment’, in Third IEEE International Conference on Pervasive Computing and Communications Workshops, Mar. 2005, pp. 15–19. doi: 10.1109/PERCOMW.2005.65. [21] M. Zoll, D. Jäck, and M. W. Vogt, ‘Evaluation of Predictive-Maintenance-as-a-Service Business Models in the Internet of Things’, in 2018 IEEE International Conference on Engineering, Technology and Innovation (ICE/ITMC), Jun. 2018, pp. 1–9. doi: 10.1109/ICE.2018.8436272. [22] S. Scholze, K. Nagorny, K. Stöbener, and D. Brückner, ‘An approach for context sensitive product extensions services’, presented at the 2017 IEEE 15th International Conference on Industrial Informatics (INDIN), IEEE, 2017, pp. 1181–1186. [23] P. Dziurzanski, S. Zhao, S. Scholze, A. Zilverberg, K. Krone, and L. S. Indrusiak, ‘Process planning and scheduling optimisation with alternative recipes’, atAutomatisierungstechnik, vol. 68, no. 2, pp. 140–147, 2020. [24] S. Scholze, A. Correia, D. Stokic, N. Nagorny, and P. Spindler, ‘Tools for Human-Product Collaborative Development of Intelligent Product Service Systems’, presented at the PRO-VE 2016 - 17th Working Conference on Virtual Enterprises, 2016. [25] SmartCLIDE Context. SmartCLIDE catalogue on GitHub. (Feb. 27, 2025). Java. Eclipse OpenSmartCLIDE. Accessed: Mar. 03, 2025. [Online]. Available: https://github.com/eclipse-opensmartclide/smartclide-context [26] Zillner, S., Bisset, D., Milano, M., Curry, E., Södergård, C., Tuikka, T., & et al. (Eds.). (2020). Strategic Research, Innovation and Deployment Agenda: AI, Data and Robotics Partnership. [27] M. B. Dias et al., ‘Sliding Autonomy for Peer-To-Peer Human-Robot Teams’, in Intelligent Autonomous Systems 10, IOS Press, 2008, pp. 332–341. doi: 10.3233/978-158603-887-8-332. 19 [28] F. Heger and S. Singh, ‘Sliding Autonomy for Complex Coordinated Multi-Robot Tasks: Analysis & Experiments’, in Robotics: Science and Systems II, Robotics: Science and Systems Foundation, Aug. 2006. doi: 10.15607/RSS.2006.II.003. [29] J. Brookshire, S. Singh, and R. Simmons, ‘Preliminary Results in Sliding Autonomy for Coordinated Teams’, in Proceedings of AAAI ’04 Spring Symposium on Interaction between Humans and Autonomous Systems over Extended Operation, 2004. [30] T. M. Roehr and Y. Shi, ‘Using a self-confidence measure for a system-initiated switch between autonomy modes’, presented at the International Symposium on Artificial Intelligence, Robotics and Automation in Space (i-SAIRAS ’10), Sapporo, Japan, 2010. [31] F. Tang, M. Mohammed, and J. Longazo, ‘Experiments of human-robot teaming under sliding autonomy’, in 2016 IEEE International Conference on Advanced Intelligent Mechatronics (AIM), Jul. 2016, pp. 113–118. doi: 10.1109/AIM.2016.7576752. [32] P. Sfair Palar, V. De Vargas Terres, and A. Schneider De Oliveira, ‘Human–Robot Interface for Embedding Sliding Adjustable Autonomy Methods’, Sensors, vol. 20, no. 20, p. 5960, Oct. 2020, doi: 10.3390/s20205960. [33] T. B. Sheridan and W. L. Verplank, ‘Human and Computer Control Of Undersea Teleoperators’, Massachusetts Institute of Technology, Cambridge, MA, 1978. [34] Torsten Hahmann, ‘Welcome to CSC384: Introduction to Artificial Intelligence’. 2011. Accessed: May 09, 2025. [Online]. Available: https://www.cs.toronto.edu/~torsten/csc384-f11/lectures/csc384f11-Lecture01Introduction.pdf [35] B. Shneiderman, ‘Human-Centered Artificial Intelligence: Reliable, Safe & Trustworthy’, Feb. 23, 2020, arXiv: arXiv:2002.04087. doi: 10.48550/arXiv.2002.04087. [36] F. Psarommatis and D. Kiritsis, ‘A hybrid Decision Support System for automating decision making in the event of defects in the era of Zero Defect Manufacturing’, Journal of Industrial Information Integration, vol. 26, p. 100263, Mar. 2022, doi: 10.1016/j.jii.2021.100263. [37] H. Suzen, ‘Hybrid Intelligence in Decision Support and Predictive Crisis Management’, Jun. 2021. [38] ‘Strategic Research, Innovation and Deployment Agenda for an AI PPP: A focal point for collaboration on Artificial Intelligence, Data and Robotics’, BDVA – euRobotics, Brussels, 2019. Accessed: May 09, 2025. [Online]. Available: https://bdva.eu/news/release-strategic-research-innovation-and-deployment-agenda-aidata-and-robotics-partnership/ [39] A. Smith, ‘Public Attitudes Toward Computer Algorithms’, Pew Research Center. Accessed: May 09, 2025. [Online]. Available: https://www.pewresearch.org/internet/2018/11/16/public-attitudes-toward-computeralgorithms/ [40] P. Cingolani and J. Alcala-Fdez, ‘jFuzzyLogic: a robust and flexible Fuzzy-Logic inference system language implementation’, in 2012 IEEE International Conference on Fuzzy Systems, Brisbane, Australia: IEEE, Jun. 2012, pp. 1–8. doi: 10.1109/FUZZIEEE.2012.6251215. [41] P. Cingolani and J. Alcalá-Fdez, ‘jFuzzyLogic: a Java Library to Design Fuzzy Logic Controllers According to the Standard for Fuzzy Control Programming’, IJCIS, vol. 6, no. Supplement 1, p. 61, 2013, doi: 10.1080/18756891.2013.818190.