On deformable object handling: multi-tool end-effector for robotized manipulation and layup of fabrics and composites

Over the past decades, robotic automation has expanded in numerous industrial sectors; however, manufacturing operations involving the manipulation of non-rigid products are mostly preserved manual. The dynamic distortion of flexible objects underlines limitations in robot cognition and dexterity. Inspired by the gaps in composites industry automation, this paper presents a novel multifunctional robot end-effector for the robotic automation of composites layup. Aiming a holistic confrontation of layup challenges, the proposed end-effector incorporates tools for (a) the manipulation of sheet materials, like composite fabrics of variable dimensions; (b) the grasping of core materials, like foam blocks; (c) the handling of peripheral tools; (d) the exertion of fitting forces; and (e) the application of resin in diverse mold geometry cavities. The meticulous design of the end-effector’s configuration allows for consistent and collision-free tool usage with no excess robot tool changes, granting even higher productivity and efficiency. The enhancement in robot dexterity and proficiency for layup operations is validated within an automotive case, where semi-automation is intended for the improvement in overall workstation’s efficiency, quality, ergonomics, and well-being.


Introduction
Over the years, businesses improve their production lines in an effort to keep up with market requirements. Starting with the mass-production era, hard automation and robotics were extensively applied as they were able to keep up with demand through their tireless repeatability [1]. Heading towards mass-personalization, Industry 4.0 and 5.0, Information and Communication Technologies (ICT) are applied for balancing cost, quality, time, and flexibility [2]. It is evident that robot agents had a significant role in those manufacturing paradigms and they will maintain it in the future through cooperative schemes besides the embodying of Artificial Intelligence (AI) [3].
Despite the benefits that robot automation brings, it is noticeable that manufacturing processes that involve deformable object handling are still largely done manually.
The particularities of deformable objects, in terms of mechanics and dynamic distortion, highlight gaps in robot dexterity and cognition where future activities must focus [4,5]. These observations and remarks are also reflected in market statistics. More specifically, the market of factory digitalization is booming will billions of euros invested in automation technologies and big data [6]. Nevertheless, for the manufacturing sectors of textiles, garments, and composites, the outsourcing of production to low-cost labor counties suggests a common practice [7].
To start shifting this notion and "de-characterizing" operations as "craftsmanship," robot dexterity and cognition for deformable object manipulation need to be enhanced [8]. This will lead to releasing "off-the-shelf" automation solutions dedicated to the related industrial sectors. Focusing on dexterity, and inspired by the challenges in the composites industry [9,10], this paper discusses the design and implementation of a multipurpose end-effector for robotized layup. The end-effector comprises a set of tools that are in disposal without any time-demanding tool changes, thus forming an "unus pro-omnibus" end-effector solution. The novelty of the proposed end-effector lies in its 1 3 tools' spatial configuration that allows for the execution of diverse operations without any collisions or occlusions. This offers a wide portfolio of operations that robot agents can perform by increasing their dexterity, thus their overall skill capabilities. The importance of reaching new frontiers in the automation of the composites process is mandatory not only from economic perspectives but also for the worker's well-being. The risks when working in the manufacturing of composite materials are widely known and well documented (e.g., skin disorders eye irritations, lung malfunctions) and rarely are limited to musculoskeletal issues due to poor ergonomics [11]. In this context, robotic solutions that limit exposure to all sorts of dangerous and strenuous operations stand out as a prerequisite in modern manufacturing.
This manuscript comprises of six sections, including this introductory section, where the remainder of the article is organized as follows. Section 2 discusses previous findings and relevant work in the areas of robot end-effector design, prioritizing those with noticeable relevance to deformable materials. Section 3 starts with an explanation of typical layout operations and continues with a set of design considerations upon specification analysis. Section 4 explains fundamental design principles and describes the implementation of the end-effector's mechatronics and control system. Section 5 discusses an industrial use case where the proposed end-effector is integrated towards its verification and evaluation. A description of the evaluation findings and results is given in Sect. 6. Finally, the paper concludes with an outlook in addition to the authors' future activities, in Sect. 7.

Relevant work
As a means of interaction with its environment, the capabilities and performance of a robot are heavily dependent on its end-effector [12]. Given the large number of material handling operations (e.g., machine tending, assembly, packaging), over the years, a lot of research has been put into designing different types and sizes of grippers, in order to grasp, manipulate, and place diverse objects. There are different approaches for classifying grippers based on their actuation mechanisms (e.g., vacuum, pneumatic, hydraulic, electric), grasping principles (e.g., friction, jaw, needle), releasing principles (e.g., pressured air, electrostatic control), or sensing principles (e.g., pressure, force) [13].
Commercial manufacturers have released gripping products that mostly perform dedicated grasping operations by prioritizing high robustness and low purchase cost. In contrast, research activities have focused on developing robot tools that incorporate additional degrees of freedom (DoFs) in an effort to improve dexterity and efficiency. Following the norm of "producing more with less" [14], high-speed robotic grippers have been developed for the grasping and in-hand manipulation of complex geometries [12], allowing for fast production line reconfiguration [15].
Other activities aim at the imitation of the living for taking advantage of what nature has perfected over millions of years. Trying to mimic human actions, various anthropomorphic grippers have made their appearance in the industrial and research community, capable of executing grasping tasks and in-hand manipulation of rigid objects using multiple degrees of freedom [12,16,17]. Inspired by biological laws, similar biomimetic soft grippers have also been proposed, as of intelligent solutions in applications like biosensors, flexible electronics, smart actuators, medicine, and surgery. Besides dexterity, another critical aspect is the design of the product-oriented components (i.e., gripper components having contact with the object). The implementation of soft robotics has shown prominent results in the reduction of control complexity through the use of advanced materials, such as shape memory alloys, active polymers, and silicone elastomers [18] that can embrace different object geometries.
Excepting hardware design, researchers have also come up with different ideas for dexterous manipulation from a standpoint of physics and fixed geometries. With the colossal increase in computational performance, even more potential has been unlocked regarding grippers' capabilities. AI-based methods and deep learning algorithms have supplemented robots' grasping capabilities in much higher rates [19]. However, as pre-discussed, the handling of non-rigid materials (NRMs) still faces challenges, due to the significant shape deformation, which needs to be taken into consideration. The automated grasping tools on NRMs are of great value for a number of industrial sectors, like composite manufacturing, food processing, textile industry, etc. [20].
Focusing on the composites industry, there are some examples of automated or semi-automated solutions. Automatic tape laying (ATL), automatic fiber placement (AFP), and resin transfer molding (RTM) technologies are the most established techniques when it comes to automated composite manufacturing [21]. These solutions, apparently, require specifically developed machines that stand in need of high capital investments. That said, another approach is the automatization of traditional pick-andplace methods, which involve the lifting, transporting, and placing of materials into molds or flat surfaces [22,23]. The main problem associated with this kind of method is the challenging handling of the involved deformable, fragile, and occasionally sticky materials. Vacuum grippers are commonly used for the automation of composite material handling, where single-sided access to the material is only available for achieving lift [24]. This kind of grippers utilizes the under pressure created by an airflow coming through a nozzle in order to retract the material 1 3 [25]. Despite the capability for performing the de-stacking of composite fabrics, this kind of grasping principle is not ideal for composites with less dense weaving. Τhe aforementioned technologies hand out also weaknesses when it comes to the handling of woven fabrics (insufficient sealing, fiber damage, etc.). Different solutions can also achieve similar results, like the Venturi principle, where compressed air is introduced into an ejector [26], and creates air pressure difference by escaping through holes [27], or the Coanda principle [25]. Another approach for contact-based one-sided grasping is based on electroadhesion (EA). By embedding EA skins on gripper fingers, the dexterity of manipulators can substantially be increased as objects can "stick" to the gripper on demand [28]. However, the performance of those methods heavily relies on the object's material properties, with fiberglass components being characterized as incompatible. The use of clamping grippers is also quite extensive when it comes to manipulating composite elements. The most common application of finger grippers requires undisturbed access to both sides of the material for optimal grasping [27]. Another application is the one of needle grippers, on which multiple needles penetrate the material under processing [29]. The main drawback of this concept is that it can harm the material itself and should not be preferred when holes or partial distancing of weaves must be avoided.
A summary of state of the art solutions among with their advantages and handicaps can be found in Table 1.

Design methodology
This section discusses the methodology behind the design of the multipurpose end-effector. For better comprehension, the section starts with an overall description of a typical layup process for clarifying what are the challenges of this manufacturing process. Then, it continues by enlisting the functionalities that are needed for the end-effector towards robotized layup.

Composites layup and industrial challenges
Composites are formed by combining a number of different materials (e.g., fiber sheets, foams, honeycombs, resins), of different properties, towards the composition of a unique material with enhanced mechanical properties [30]. Their manufacturing involves complex operations that even with today's advanced technologies is considered "craftsmanship." Typically, the manufacturing of a composite prerequires the existence of mold(s) that formulate the shape of the component in question through "negative geometry" cavities. Different techniques are available, with most of them following a three-stage process, with the stages being (a) cutting/forming of fiber fabrics [31], (b) lamination of plies, and (c) polymerization of the laminate [32]. The stages with the greatest automation potential are the first two, since those involve numerous material handling operations. During the cutting process, the fiber fabrics are shaped into the desired geometries by trimming pieces out of continuous (rolled) fabric stocks. This can either take place manually with specialized tools or computer numerical controlled (CNC) laser cutting machines. In both cases, machine or workstation tending presents high repeatability of non-added value operations.
The second stage, as illustrated in Fig. 1a, involves the stacking of the different laminae where the technician(s) thoroughly overlaps all required layers at the defined positions and orientations. Most frequently, a core is also intervened, in order to improve the component's stiffness and mechanical properties without noteworthy weight addition. Common core materials, used in the industry, are polyvinyl chloride (PVC), polyurethane or polyethylene terephthalate (PET) foams, plywood, or aramid paper. For proceeding into the final stage, it is important to infuse resin between the layers. This can take place in between the placement of each layer by spreading epoxy using brushes or rollers. The technician applies pressure on the working surface, by rolling and squeezing evenly, to remove any gaps or air bubbles in between layers. Alternative tools with different profiles are used in accordance with the geometry of the laminated area (i.e., flat roller for flat surface, chamfer roller for internal corners, dabber tool for tight internal cavities), as presented in Fig. 1b. In overall, this is a consuming process in terms of time and human resources and the final product quality heavily depends on the skills of the involved personnel. The manual application of resin can be circumvented via "resin infusion" where resin is injected with high pressure on dedicated molds by filling all layer gaps. It is noteworthy that technicians may also use stapleguns or spray glue in order to secure the fabric pieces in place until in between layup steps. Finally, the last stage is polymerization where the composite obtains its mechanical properties by curing at a set temperature until its resin reacts and hardens to a solid state.
A holistic confrontation for automating layup operations necessities for robots to use a plethora of tools. For the rest of the manuscript, it is clarified that the design activities focus on automating the lamination layup and roller-based resin spreading, considering that it is the process with the heaviest material handling workload. It involves miscellaneous objects such as fabrics and cores and implies the application of insertion or resin-spreading forces over diverse geometries. The design's complexity rises when additional tools need to be manipulated like stable guns or air glue sprayers.

Functional requirements for end-effector design
Automated solutions of composite manufacturing are limited, especially during the lamination phase of hand layup, where most challenges arise, owing to the diversity of components' dimensions and complex molding geometries.
In more detail, and as discussed in the previous subsection, such solutions require a dexterous robotic configuration able to (a) grasp and handle limp materials, like laminae; (b) grip firm ones, such as foam blocks; (c) layup surfaces, by rolling and squeezing positioned materials; (d) fix positioning of layers by stapling or spay-gluing them; and provisionally (e) de-stalking fabric layers prior grasping. Aiming for having a single dexterous end-effector that can tackle all aforementioned tasks, not only would conserve resources but will also increase efficiency and productivity by drastically reducing deadtimes for tool-changing. Based on axiomatic design [33], the functional requirements that can be extracted are summarized in the following attributes: • Grasping and picking mechanism for fabrics: For fabrics unfolded on a plane, there is usually accessibility only for one-sided grasping. Thus, the usage of needle grippers, EA skins, or vacuum suction cups are considered the most effective methods. Depending on the fabric weaving, needles are preferable in the cases where vacuum seal cannot be achieved, plus spotless material integrity is not a priority. EA skin can resolve both limitations; however, it pre-requires compatible materials. • De-stacking: In cases where fabric laminates are stacked prior to their transferring, vacuum or EA grasping is preferred over needles for de-stacking. By properly adjusting vacuum pressure, the lifting of multiple layers can be avoided. • Grasping point distance reconfiguration: The fabrics, used in the industry, come in all shapes and sizes. Therefore, it is desired to actively control the distance between grasping points. The same functionality also provides the ability to in-hand manipulate the fabric for controlling its stretching or curvature. • Pick-up mechanism for foams: Materials like foam blocks are commonly used as core materials and, therefore, parallel friction-based gripping is ideal for most core shapes. • Layup mechanism: Based on the tooling that is currently used in the industry for the manual process, customshaped rollers and pressing tools need to be deployed for the layup on flat or corner surfaces.
The inclusion of all those functionalities into a unified tooling system sets an engineering challenge. Successful design suggests that none of the integrated functionalities will handicap the operability of others. Henceforth, intelligent tool spatial arrangement and provisionally exploitation of the same mechanisms for more than one functionality are of substantial importance.

Implementation
This section presents the design and implementation of the proposed end-effector for automated layup. The section starts with a description of the overall device and its specific sub-mechanisms. Then, it continues with an analysis of the implemented control framework and its interfacing with service-oriented robotic systems.

Multipurpose layup end-effector design
Inspired by the challenges in composites manufacturing and the functional requirements that were described in Sect. 3.2, the end-effector of Fig. 2 is designed. Figure 3 provides a visual understanding of the device kinematics, with the main actuation mechanisms employed being a stepper motor for The proposed design couples all four operations in a single cost-effective device, by involving the following sub-mechanisms: • Parallel grasping mechanism: A brushless servomotor, with an integrated planetary gearbox, drives a leadscrew mechanism. By splitting the drive shaft into clockwise (CW) and counter-CW, two holding pads can mirrorlike "open" or "close" with the same power input by the intermediate timing-belt powertrain. The pads are designed for achieving friction-based parallel gripping (Fig. 4b), for core materials, or parallel grasping of tools, such as staple guns, using additional alignment features. • Composite fabric grippers: A pair of single-acting, pneumatic needle grippers penetrate between the composite weaves and secure it for transferring (Fig. 4a). The distance between the needle grippers, or grasping points, can be laterally adjusted with the use of the leadscrew drive mechanism of the parallel gripper. This adjustment of the grasping point distance is valuable for in-hand manipulation or grasping of composites of different widths. • Layup rollers: A flat-surface (Fig. 4d) and an edgesurface roller (Fig. 4e) can independently be deployed for foam handling via parallel grasping mechanism, c auxiliary tool handling via parallel grasping mechanism, d layup roller for flat surfaces, e layup roller for internal curves or curved cavities, and f dabber tool for pressing at tight corners evenly distributing resin during the lamination process. A double-acting pneumatic cylinder is responsible for lowering and retracting each tool. By adjusting the air pressure of actuation air, the stiffness of the tool can be finetuned. • Dabber tool: A dabber tool is attached to one end of the end-effector to statically flatten layers in challenging mold geometries, like corner edges (Fig. 4f). • Grasping of peripheral tools: Actuated by the parallel grasping mechanism, the two holding pads have custom alignment features. Those are chamfered grooves that secure the positioning of the manipulated tools as long as the former ones are equipped with compatible grasping adapters. In this sense, the robot is able to grasp stapling guns (Fig. 4c) or spray gluing guns for performing additional tasks of interest.
The device of Fig. 2 has a total length of approx. 500 mm and can manipulate fabrics, with a maximum opening of 450 mm, as well as foam blocks of up to 350 mm width. A center-out or parametric design approach has been adopted, meaning that with small modifications, the device can be expanded to greater length dimensions by elongating the lead screws, bearing track, and the supporting rods. The complete structure weighs around 4.8 kg and can be integrated on most commercial low payload robots. Another important feature that is affected by the lead-screw mechanism is the speed of the parallel grasping mechanism. For reference, the brushless direct current servomotor (by Faulhaber) with the selected planetary gearhead, driving belt and leadscrew powertrain transmission is capable of completing a full-length stroke in under 6 s. The gear ratio, thus cycle time, can be adjusted by replacing the powertrain components with an effect on the parallel grasping force as well. In the integrated solution, more than 90% of the required stroke is performed within a robot trajectory (i.e., pre-pick and post-release phases), whereas for the actual grasping or release actions, only a few millimeters of stroke are sufficient.
To resolve de-stacking problems, a variation of the proposed device has also been designed. This replaces the needle grippers with a combination of suction caps and angular finger grippers. More specifically, the suction cups are vertically transposed by double acting pneumatic cylinders. As illustrated in Fig. 5, for de-stacking a fabric, a three-phase operation is foreseen as follows: (a) the robot approaches the fabric and the cylinders lower the suction cups that generate vacuum, (b) the cylinders raise the suction cups for achieving a clearance between the grasped fabric and the stack, and finally, (c) two angular grippers secure the fabric for transferring. Comparably to the original design, the distancing of the grasping points is adjusted by the parallel grasping mechanism, whereas the grasping of peripheral tools is available with the same pads that are mounted on each angular gripper's body.

Control
The control of the proposed end-effector is built on top of a multi-layer framework that involves different communication protocols and practices for the operation of all actuation and sensing mechanisms (Fig. 7). For ensuring compatibility with the rest of the system, the end-effector is controlled via Robot Operating System (ROS) [34], through a hardware interface that forms the top controlling layer. It is responsible for wrapping all commands and machine state messages, based on the actuator manufacturer's control map, into standardized ROS messages. This allows for the state, triggering, position or velocity variables to be read, updated, and written accordingly. For facilitating users of less advanced programming skills, two user interfaces have been implemented. The first one, as depicted in Fig. 6a, can be used as a standalone application interface, whereas the second one, as presented in Fig. 6b, allows for direct device control through the robotic system's digital twin. Overall, gripper is fully integrated through MoveIt and can be simulated both in RViz and Gazebo. Subsequently, within a digital twin or simulation model, the gripper can interact with the environment, perform grasping tasks, and a physicsbased simulation of its behavior can be observed.
The following layer includes each distinct communication protocol that links the hardware interface with the end-effector's mechatronics. For the lateral motor actuation mechanism, the messages are communicated to the motor controllers through Controller Area Network (CAN-bus). A CAN to USB adapter is used for establishing the bus between the robotic cell's desktop and the motor's controller. The exchanged messages (composed of the identifier bits, followed by 8bytes of user data) allow for different SDOs (Service Data Objects) that support the required monitoring or control modes, with the most important ones to be the (a) homing mode, for calibrating motor's position; (b) profile position mode, for controlling motor's position; and (c) profile velocity mode, for controlling the motor's velocity. For controlling the pneumatic actuators, a custom layer for exchanging data packages with checksums (through USB) is built on top of UART. Through this layer, a Controllino I/O module is operated by the same robotic cell's desktop for triggering or reading digital or analog signals. Subsequently, the configured ROS messages can independently actuate the wired pneumatic valves. For the implemented end-effector design, which involves two needle grippers and two rollers, three 5/2 solenoid valves are engaged.
Elaborating on the calibration of the motorized parallel actuation mechanism, a position switch is placed near the middle of the frame. During a calibration operation, the leadscrew closes the parallel gripper mechanism until one of the two needle gripper holding brackets touches the position switch. Upon contact, a signal is sent to the motor's controller for completing the calibration operation by zeroing the current opening value (Fig. 7).
It is clarified that the Controllino I/O module can be replaced by a PLC unit that will communicate with the rest of the system through ROS/OPCUA or ROS/MQTT.

Case study and results
The multi-purpose end-effector has been developed to resolve the (semi-)automation of composites manufacturing by addressing a number of operations related to skin and core material handling. The proposed device is integrated on a robotic cell that is intended to optimize the working environment conditions and the key performance indicators of a composites manufacturing workstation. The workstation deals with the assembly of automotive lightweight panels for large vehicles via resin infusion. In more detail, the process can be described as a sequential set of pick, place, and layup operations, where fabrics or cores, of different shapes and sizes, are transferred from a buffer to the mold. Appropriate placement of the components is crucial for the mechanical properties and the quality of the final product. Upon their placement, the components must be flattened and pressed against the mold for ensuring that there are no wrinkles or gaps. Regularly, the stacked fabrics are secured with staples for guaranteeing that they will not be shifted during the upcoming layup step(s).
Due to the large size of the panels (and their respective molds), there are areas with inadequate or no accessibility for the operators. This creates ergonomic challenges, besides quality fluctuations, since operators have to climb and crawl on the mold for diverse intervening actions. Ergonomics is further affected by the repeatability of the process (i.e., large number of fabric layers and pieces) as well as the limited, yet actual, exposure to chemicals. With the objective of improving ergonomics, well-being, and performance, a robotic cell based on a 14-degree of freedom dual-arm manipulator is implemented (Fig. 8). The proposed end-effector equips the arms, allowing for the manipulation of composite fabrics, foams, and supplementary tools. More details about the envisioned solution as well as the planning principles for overcoming the materials' deformities during co-manipulation can be found in [35,36].
As a backbone component of the integrated solution, the end-effector was utilized and evaluated for all intended applications. As presented in Fig. 9, testing involved (a) fabric transferring for different types and dimensions, (b) foam block transferring, (c) layup with curved or (d) flat rollers and (e) application of forces with dabber tool, and (f) stapling of layers.

Results and discussion
The proposed design not only must be able to perform the intended operations but also perform them at certain repeatability levels. Through a series of tests, the end-effector showed great compliance with the defined specifications and demonstrated high performance in terms of repeatability. In overall, the following results are worth noting: • The adjacent needle grippers with the parallel mechanism project sufficient grasping capacity and fitness to multiple fabric variances. The end-effector can grasp and manipulate fabrics of multiple variations, different textures, and thicknesses. For fabrics of small to medium width, the end-effector is sufficient, whereas for fabrics of larger width, the second arm (with an identical endeffector) can be engaged (Fig. 10). In-hand manipulation in the form of fabric stretching is also available through the parallel actuation mechanism. • The needle grippers can disorder the composite's weaving by distancing or curving the fibers near each grasping point. For most applications, this kind of defect can be neglected; however, for space or aviation industries, the strict quality standards might propose alternative grasping principles. • For the foam blocks, the manipulation is performed through friction-based parallel gripping. The payload capacity is dependent on the friction achieved by the normal force that is generated by the leadscrew mechanism. Testing proved that the payload capacity is sufficient for all longitudinal foam block components of interest (that are within the payload capabilities of the robot arm as well). Foams used in the industry, largely vary in size, density, compression and surface properties, etc. In this sense, for foams with less porous surfaces and small thickness sizes, the best approach would be to use a needle-based grasping approach. An advantage of needle gripping systems is the precise control of the contact area in combination with the lack of object compression during manipulation. On the other hand, a friction-based approach would be more suitable in foams Fig. 9 Evaluation of functionalities: a fiberglass transferring, b core foam transferring, c chamfer roller layup, d flat roller layup, e dabber tool force application, and f stapling fabrics in position (tool replica), g needle-based foam transferring, and h dual arm fabric transferring of high porosity, rough textures, or irregular shapes. It relies on created sufficient pressure force between the gripper and the foam object to maintain secure grip. • Referring to the force application and layup, both rollers and the dabber tool proved their fitness for the selected mold cavities. No incompatibility issues were identified; however, the geometry of the curved roller could be optimized for covering the targeted areas after a smaller number of robot trajectories.
The evaluation process included some biased trials where it was attempted to unveil any design handicaps or areas that can be further improved. The most important one refers to the slippage of the foam blocks during human robot co-manipulation. In detail, when the operator pulled, raised, or lowered the foam blocks in an "aggressive" fashion, slippage could occur in terms of grasping positioning or orientation. This can be circumvented by inversing the "peg in hole" alignment features of auxiliary tooling, by locating the pegs/pins on the parallel fingers. However, this is not suitable for less deformable foam blocks and there is a high chance that improper operator handling can lead to foam breaking instead of (defect free) grasping slippage. Another improvement could be on the material of the rollers that currently are made from ertalon. A softer and deformable, yet stiff, material could significantly reduce the number of robot trajectories through better compensation of the mold's geometries by the rollers. Finally, the prototype needs to be equipped with sensors for monitoring the status of the pneumatic actuators as well as grasped fabrics. For the actuators, market available, magnetic build-in sensors can be installed for acknowledging the stroke status, whereas an optical or laser sensor, between the two grasping points, is sufficient for identifying any fabric or foam drops. In the present study, no explicit comparison between the quality of the final composite obtained from the use of the proposed device and human labor takes place. Focus is put on addressing the requirement specifications of the end-users, mainly focusing on the material placement on the mold. Based on the performance results of the layup use-case described in chapter 5, the specifications set by the end-users were successfully achieved, indicating that characterization experiments would yield acceptable results in a direct comparison of product quality between automated and labor process. Moreover, all quality requirements up to the phase of composite layup were met and accepted, according to end-users' feedback. It is worth noting that even some small defects, created by the needles, on the fiberglass material was deemed insignificant, overall, by the end-user.
Summarizing and referring to the objectives of the complete robotic solution, the automation of the layup manufacturing is feasible by incorporating a plethora of tools for increasing the dexterity and cognition of robots. The proposed end-effector design managed to offer the required dexterity, at least for the proposed case study, by presenting the necessary grasping and manipulation skills to the robot agents. Upon integration, the handling of both skin and core components was validated, meaning that the robot agents can undertake activities that originally were allocated to human operators. In the long-term, this can contribute to an improvement of the overall product quality, performance metrics, and well-being. The introduction of robots as well as monitoring and reasoning algorithms can reduce the dependency of composites manufacturing on human operators by identifying or preventing errors or fluctuations. Moreover, the reduced operator full time equivalent for such workstations suggests that ergonomics can be improved through job-rotation and even less exposure to chemicals.

Conclusions
This work discusses the design and implementation of a novel multi-tool end-effector for robotized composites layup. Inspired by the gaps in composites industry automation, this multifunctional device aims for a holistic confrontation of layup challenges in a time and cost-efficient manner, without the need for intermediate tool changes. The endeffector is equipped with tools for (a) the manipulation of fabric materials, (b) the grasping of core materials, (c) the handling of peripheral tools, (d) the exertion of fitting forces, and (e) the application of resin in diverse mold geometry cavities. Design variations are also proposed for addressing additional industrial problems, like de-stacking or defect-free material grasping. The selected spatial arrangement of tools in combination with their actuation mechanisms allows for collision-free tool usage without any confrontations. Last but not least, the implementation of a ROS driver, that interfaces the actuators' controlling and monitoring functionalities with service-oriented robotic systems, proved to be an advantage.
The end-effector was integrated and validated on a robotic system for the enhancement of working conditions and performance of an automotive industry workstation. By offering the required dexterity for the handling of all core and skin materials of interest, besides their lay-up, the end-effector contributed to the holistic semi-automation of a process that traditionally is preserved manual. This paper aims to inspire engineers for designing and implementing handling tools that can increase robots' dexterity in sectors that present many challenges for robotics. The findings of this work and the validation of the discussed concepts can benefit stakeholders from academia and industry that aim to automate scenarios from composites or textiles industry.
Future work of the authors includes the enhancement of this prototype through its "clothing" in safety skins for even closer human robot coexistence. Focus will be given on enhancing monitoring and perception capabilities by integrating multisensory systems for object detection and error handling. Moreover, different methods for the manufacturing of layup rollers, in terms of materials and geometries, will be investigated for greater efficiency during layup. Finally, the implementation and validation of the device on scenarios deriving from other industries will further support its universal design principles.
Author contribution All authors contributed to the study conception and design. Conceptualization of the device was performed by Dionisis Andronas and Giorgos Papadopoulos. Implementation of the device was performed by Giorgos Papadopoulos, Dionisis Andronas, and Emmanouil Kampourakis. Control architecture, algorithms, and software were designed and implemented by Nikolaos Theodoropoulos, Emmanouil Kampourakis, and Panagiotis Kotsaris. Activities management and technical overview were covered by Dionisis Andronas and Sotiris Makris. Material preparation, data collection, and analysis were performed by Giorgos Papadopoulos, Nikolaos Theodoropoulos, and Panagiotis Kotsaris. The first draft of the manuscript was written by all authors. All authors read and approved the final manuscript.
Funding Open access funding provided by HEAL-Link Greece. This manuscript has been partially supported by the Project "MERGING -Manipulation Enhancement through Robotic Guidance and Intelligent Novel Grippers," funded by the European Commission under the Horizon 2020 Research and Innovation Programme (H2020-DT-FoF-12-2019) with grant agreement ID: 869963.

Competing interests The authors declare no competing interests.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.