Human-Centric Robotic Solution for Motor and Gearbox Assembly: An Industry 5.0 Pilot Study

Abstract

The automotive industry is one of the most automatized industries, employing more than one million robots worldwide. Although several steps in car production are completely automated, many steps are still carried out by operators, especially in tasks requiring high dexterity. Additionally, customization and deployability are still pending issues in this industry, where a real collaboration between robots and operators would increase the reconfigurability of the assembly lines. This paper presents an innovative robotic cell focused on the motor and gearbox assembly, including collaborative industrial robots and autonomous mobile manipulators along the different assembly stations. The design also incorporates a human-centered approach, with an enhanced human interface to facilitate the interaction with operators with the complete robotic cell. The proposed approach has been deployed and validated on a real automotive industrial scenario, obtaining promising metrics and results.

1. Introduction

The introduction of robotics in the industry has been a reality since the robot revolution in manufacturing [1] in the 1970s. During the last decades, robots have expanded throughout multiple sectors, automating a large amount of tasks and changing the design of the assembly lines. Within all the industrial sectors, the automotive one has been a crucial actor in the deployment of robots in assembly lines, with more than one million robots installed worldwide [2]. Even so, this robotization is not equally distributed among the different assembly steps of a car. Although car body construction, welding, and painting are highly automated, there are still fully manual assembly steps, such as the final assembly line of cars and engines.

In this sense, the Fourth Industrial Revolution or Industry 4.0 [3] put on the fore several technologies that will enable the automation of assembly steps currently completely carried out by operators. From the use of mobile robotics to the adoption of advanced human-machine interfaces and wearable devices, many of the technologies posed by this paradigm facilitate the creation of advanced assembly stations where connectivity and operator assistance are central topics. The recent Industry 5.0 [4] extends this concept, emphasizing a sustainable, human-centric, and resilient industry where the worker’s well-being is a central element of the production process. Even so, current implementations still rely on classic robotics where robots and operators work separately, without exploiting the potential of joint work between humans and robots.

This paper presents a human-centric industrial pilot for the assembly of combustion engines where robots assist operators during the manufacturing process in a shared workspace. The main contribution of the presented work is the design and implementation of a collaborative assembly line where industrial and mobile manipulators assist operators in the assembly of motors. Additionally, the human-centric approach integrates augmented reality to facilitate interaction with the system and provide feedback about the status of the process thanks to the digital twin of the whole cell. The presented pilot has been deployed and validated in a real industrial car manufacturing environment, obtaining promising results and feedback from operators.

2. Related Work

Advances in technology are continuously evolving the configuration of factories and assembly lines. The recent Industry 5.0 paradigm aims to drive this evolution towards a human-centric approach where operators and robots collaborate in production lines, enhancing worker’s skills through new robotics technologies. In this sense, Zafar et al. [5] analyse the synergies between collaborative robotics, digital twins, and augmentation in Industry 5.0, highlighting the large variety of application possibilities of the paradigm. Among all the key technologies, collaborative robotics is one of the most studied fields [6,7] due to its clear impact on creating a human-centric production line. Several works have applied the Industry 5.0 paradigm to different applications such as robotized composite fuselage panel inspection [8] or flexible assembly and repair mechatronics line [9].

Focusing on the automotive industry, car manufacturing is one of the most robotized sectors, as well as one of the most important driving actors in the development of robotics over the last decades. Even so, this robotization is concentrated on the initial car manufacturing steps like car body assembly and welding. On the contrary, the final assembly line mainly employs operators in the majority of the stations. In this sense, the application of collaborative and mobile robotics can help to break this barrier, as pointed out by the work of Land [10], allowing the automation of tasks currently carried out by humans.

Even so, the introduction of collaborative robotics in automotive assembly lines raises multiple challenges, especially regarding safety and quality in continuously moving lines or handling heavy parts. Gopinath et al. [11] implemented an automotive case study focused on a collaborative panel assembly operation with large industrial robots, defining a layout based on the hazards identified during the risk assessment. Complementarily, Heydaryan et al. [12] propose the use of hierarchical task analysis (HTA) to simulate and evaluate the performance of collaborative assembly tasks, testing the approach on an automotive brake disc assembly operation.

Another key aspect of the deployment of robotics on assembly lines is the human-robot interaction. The employment of displays and standard devices is impractical in assembly lines due to the interference caused in the process. To overcome this issue, the employment of wearable devices significantly decreases this interference, allowing the implementation of new paradigms like augmented reality. In this sense, the automotive industry has been a recurrent testbed of this kind of technology [13]. Michalos et al. [14] propose the combination of high-payload industrial robots with human capabilities under a fenceless environment, validating the approach in two automotive scenarios, the loading of the rear axle and the assembly of the two-wheel groups. Wulff et al. [15] apply augmented reality for iterative robot program optimization in robot-automated series production, showing the viability of the approach in sealing application operations.

After revising the literature, the efforts to evolve the current assembly lines to a human-centric approach posed by the Industry 5.0 paradigm are clear. Nevertheless, most of the references focus on the application of some of the technologies and concepts of the paradigm, overlooking several other aspects of a human-centric model. The presented work tries to cover most of the relevant technologies necessary for the implementation of a human-centric assembly line for the automotive industry, integrating industrial and mobile manipulators, safety standards for the creation of a collaborative environment, and enhanced human-robot interaction through augmentation.

3. Proposed Approach

This paper presents an industrial pilot for the assembly of combustion engines based on the Industry 5.0 premises where robots and operators share the workspace. The approach seeks to implement a human-centric assembly station, extending autonomy through the real collaboration of robots and humans and allowing the execution of robot-assisted assembly tasks in a fenceless environment. The approach also expands the operator’s perception of the assembly station through augmentation, offering real-time feedback on the cell’s status based on wearable devices and the digital twin of the line. The main contribution of the approach is the implementation of a human-centric assembly station based on the guidelines posed by the Industry 5.0 paradigm, integrating most of the relevant technologies and paying special attention to key aspects such as safety standards and augmented operator capabilities.

Focusing on combustion engine manufacturing, the current assembly lines have some specific features that are also common in car manufacturing:

• During the assembly of the engine, it is necessary to manipulate several heavy parts, such as the gearbox, where operators make use of weightless manipulators or cranes to load the parts and attach them to the engine body.
• The engine is transported along the assembly line through the different stations where operators carry out the different assembly steps. Although the majority of the assembly tasks are performed in a stationary way, some of the actions are conducted in movement while the engine is being transported between stations.
• Despite some of the production steps being fully automated, especially those related to the engine body fabrication and machining, many assembly phases involving manipulating small parts are still manual. Therefore, there is a clear separation between robots and operators along the assembly line. This lack of collaboration between robots and humans is also common in the automotive industry.

Based on these premises, the proposed approach poses three main features as key elements of the human-centric engine assembly pilot to overcome the current limitations of the lines:

• Creation of a collaborative environment where industrial robots and operators cooperate in a safe and fenceless environment following the current safety standards.
• Enhanced operator capabilities to interact with the robotic cell and receive real-time feedback on the assembly line status through the use of wearable devices, augmented reality, and digital twins.
• Capability to automate different operations around the production line as well as on moving parts employing mobile manipulators to provide flexibility to the manufacturing process.

The next lines provide further information on the layout of the proposed human-centric engine assembly line.

Robotic Cell Layout

As stated previously, the main idea of the presented work is to design a robot-assisted assembly line where operators’ capabilities are enhanced by human-robot collaboration. Specifically, the proposed pilot presents the following layout, see Figure 1, focused on the assembly of the gearbox and engine cover, including the automated quality inspection. The robotic cell is divided into three different assembly stations:

• The collaborative high-payload part assembly station where operators interact with a high-payload industrial robot to handle a gearbox and connect it to the main engine body. Additionally, operators place the engine cover on the top part, positioning the bolts that link it to the main body.
• The robotized mobile screw fastening station delivers the engine to the next assembly area, leveraging this transportation time to fasten the screws of the engine cover. The main purpose of the station is to exploit idle transportation time to carry out operations on the engine while it is moving, applying the capabilities of mobile manipulators to fasten the engine cover screws during the conveyance and saving a stationary assembly station on the line.
• The robotized quality inspection station verifies the state of the previous assembly steps using artificial vision. The station takes advantage of the use of mobile manipulators to add flexibility due to the high degree of mobility of the robotic platform, allowing the verification of multiple engine elements and even parts of the same station.

This layout allows the implementation of a safe and fenceless environment where operators and robots interact and collaborate, adding the capability to automate operations on moving parts during transportation through mobile manipulators.

4. Assembly Stations

As presented in the previous section, the proposed layout is composed of three different stations: The collaborative high-payload part assembly, robotized mobile screw fastening, and robotized quality inspection stations. The following paragraphs provide further details of each of the stations.

4.1. Collaborative High-Payload Part Assembly

The first collaborative high-payload part assembly station tackles the manipulation of heavy parts by operators, specifically the assembly of a gearbox in the main engine body. This gearbox weighs more than 35 kg, which can lead to injuries due to the manipulation of heavy parts. To overcome such ergonomy and strain issues, the collaborative high-payload part assembly station introduces a human-robot collaboration paradigm towards the assembly of high-payload parts based on hand guidance and augmented reality (AR) technologies. In particular, hand guiding allows precise and effortless gearbox manipulation and positioning, enhanced by AR to guide operators along the assembly task to ensure its final quality.

This operation is composed of three steps. In the first step, the operator transports the gearbox from the kitting pallets to the assembly area. In the second step, the main axle of the engine body is aligned with the gearbox to perform the screwing of the parts. This alignment is a very sensitive task as it requires a non-specific force, strong enough to rotate the main axle of the gearbox but also weak enough to avoid removing the motor from its fixture inside the assembly area. In the final step, the operator uses a screwdriver to connect the two parts, tightly attaching the gearbox to the main axle.

From the hardware point of view, the central element of the station is a high-payload collaborative Comau AURA [16] robot, which is in charge of handling the heavy assembly parts. As mentioned previously, in the initial phase the robot needs to transport the required parts from the kitting pallets to the assembly area. To this end, two custom gripping tools have been designed based on the geometry of the engine body and gearbox, as presented in [17]. Both motor and gearbox grippers, see Figure 2, are based on a set of high-payload pneumatic actuators installed on a steel flange. Each pneumatic actuator is equipped with a still pin matching the selected grasping points of the part. To ensure the high-payload ability of the grippers, extra steel pins are used to distribute the grasping force to multiple points of the grippers.

Once both the motor and gearbox are on the assembly area, operators must align the main axle of the engine with the gearbox to perform the final screwing of the parts. The alignment is a very delicate task, as any sudden movement might damage the parts. To facilitate the process, hand-guiding techniques allow operators to move the high-payload collaborative robot manually to the desired assembly position. The hand-guiding functionality of the robot is based on the installation of a custom-designed joystick on the end effector of the robot, see Figure 3. This joystick is equipped with (a) one force sensor able to measure the forces applied by the operators and send it to the corresponding robot controller to generate motion commands, (b) one enabling button for the operator to activate the hand-guiding functionality, and (c) one emergency button to stop the robot in emergencies.

Additionally, an augmented reality application has been developed to ensure the correct execution of the task. Specifically, the AR application guides the operators on the location of bolts that must be assembled on both the engine and the gearbox, as shown in Figure 4. The location of the bolts is presented inside the virtual world of the AR application using holograms, assisting operators and avoiding assembly errors. After the execution of the screwing tasks, the operator is able to inform the production system about the completion of the tasks using a virtual button and several custom-designed interfaces for the connection of the AR application with the production system.

The implementation of the AR application is based on the Microsoft HoloLens [18] device and Unity3D [19] platform in combination with the Mixed Reality Toolkit [20] (MRTK). MRTK allowed designing the appearance of the virtual interfaces, representing holographic virtual objects such as bolts, and detecting various real-world components such as operator hands or calibration markers. Additionally, the integration with the complete cell system relies on ROSbridge using the ROS# library [21], facilitating the communication between ROS and the AR application.

This combination of hand-guiding and AR technologies enables the implementation of a global assistance, emphasising the human-centric approach of the station.

4.2. Robotized Mobile Screw Fastening

This second robotized mobile screw fastening station tries to exploit idle transportation times to carry out operations on the engine while it moves along the assembly line. Specifically, once the operator assembles the gearbox to the engine body in the collaborative high-payload part assembly station, the next step of the process requires installing the plastic cover of the motor. To this end, the operator initially positions the cover on the top of the engine and places the bolts in their respective holes, bolts that must be fastened using an electric screwdriver. To save time on this last step, assembly lines can opt for the option of screwing the bolts on the move while the engine is transported to the next assembly station. Even so, this kind of task is a vector of stress for operators as the continuous movement along the line can cause both physical injuries as well as mental strain. Therefore, the designed pilot proposes the use of mobile manipulators to perform the screw fastening operation, thus automating one of the mentally demanding tasks and encouraging a human-centric paradigm.

Specifically, the robotized mobile screw fastening station presents an autonomous mobile manipulator, see Figure 5, equipped with an automatic screwdriver and a vision system to track parts and work on them while they are being transported. The specifications of the mobile manipulator are listed below:

• Omnidirectional mobile platform with mechanum wheels [22] able to move at a speed of up to 1.0 m/s. The platform is equipped with two Kuka LBR iiwa 7 [23] manipulators in a dual-arm configuration, although a single arm is used for the presented process.
• Industrial IDS [24] UI-5240CP monochrome camera and an LED lighting system to ensure the illumination conditions and minimize detection errors.
• OnRobot multifunctional screwdriver [25] incorporating torque control as well as intelligent error detection during the screwing process.
• From the software point of view, the mobile platform includes the standard ROS Navigation Stack [26] as well as a custom mobile Visual Servoing architecture presented in this paper [27].

Based on the previously presented hardware and software specifications, the autonomous mobile platform can fasten the screws of the engine while it is transported along a conveyor at a speed of 100 mm/s as shown in Figure 6. This approach represents an advancement in two key aspects of future assembly lines: The capability to automate assembly tasks on moving parts and the reduction of mental strain on operators following the Industry 5.0 premises.

4.3. Robotized Quality Inspection

The last robotized quality inspection station focuses on verifying the assembly process, a crucial step in most industrial sectors such as the automotive one. Although automatized quality inspection has successfully been applied over the last decades, the current solutions lack a human-centric perspective. The absence of feedback from quality inspection systems to operators can lead to delays in the process, especially in those assembly lines where fabrication defects must be fixed by these same operators. The proposed approach tries to overcome these limitations by merging robotized quality inspection with augmented reality (AR) technologies to enable seamless and real-time feedback from the inspection system to operators.

Specifically, the proposed robotized quality inspection station includes a mobile inspection unit, composed of a mobile manipulator with an embedded vision device, which offers flexibility to carry out inspection activities along the assembly. In particular, the quality of the assembly of the engine and gearbox is conducted through a RC Visard 65 (Roboception, Munich, Germany) stereo camera [28] mounted on a Comau mobile manipulator, see Figure 7. This custom platform, composed of a Comau Agile 1500 AGV [29] and the Racer 5 Cobot [30], provides the capability to autonomously navigate around the shopfloor executing quality inspection tasks along the assembly line.

Additionally, the station is enhanced by an AR application that operators use to interact with the quality inspection system. Once the developed quality inspection module evaluates the correct installation of parts (screws, valve pipes, and connectors) on the engine and gearbox assembly, the operator’s intervention is requested whenever an incorrect assembly is detected. Initially, the AR application provides the quality inspection results, see Figure 8a, as well as corrective instructions to operators in the form of holograms, Figure 8b. Whenever a faulty part is detected, operators approach the engine and follow the instructions provided by the AR application, using virtual buttons to interact with the system. Finally, the operator can validate that the proposed corrective actions have been completed, informing the system that the product is ready for the next stage of the production line as shown in Figure 8c.

Summarizing, the robotized quality inspection station enhances the traditional automatized quality inspection through the use of augmented reality, assisting operators whenever corrective tasks are required on the engine assembly process.

5. Safety Implementation

A key component for the implementation of the proposed human-centric robotic solution is the design and implementation of a safety concept to ensure the well-being and security of human operators when working in shared workspaces with robotic resources. As the presented assembly cell includes multiple robots, both high-payload industrial and mobile manipulators, as well as collaborative tasks between robots and humans, it is necessary to define a layout that considers all these previous aspects. The safety layout specifically designed for the pilot is illustrated in Figure 9.

The layout is divided into two main safety areas, the high-payload robot area and the mobile manipulator area, where the screw fastening and the quality inspecion tasks are perfomed. Additionally, the set of safety elements and interfaces are included along the assembly line. The key components for the implementation of the robotic cell safety concept are presented below:

• Programmable Logic Controller (PLC): The safety PLC is the central element of the safety system and it is fully integrated with all safety hardware devices within the robotic cell as well as all the robotic resources and conveyor. The main safety logic is implemented within this PLC to enhance performance and ensure the reliable and consistent operation of the process.
• Light barriers: The safety concept of the high-payload robot area is designed around the installation of one safety light barrier (Figure 10a), complemented by two strategically placed mirrors to ensure comprehensive coverage of the high-payload robot’s operational perimeter. The placement of the light barrier and mirrors is determined by the safety formula based on EN ISO 13855 [31]. The safety light barrier are capable of detecting the presence of human operators and automatically stopping the assembly process if an operator enters the detection zone, ensuring a safe human-robot collaboration environment.
• Emergency and reset buttons: After analysing the layout of the robotic cell as well as the human and robotic tasks towards motor and gearbox parts’ assembly, six emergency stop buttons (Figure 10b) and one reset button have been installed within the cell layout. The emergency buttons stop the assembly process whenever operators perceive any emergency situation. On the contrary, the reset button enables operators to resume the assembly process after any emergency stop, safely resuming the process by pushing the button. All these buttons are integrated with the safety PLC of the robotic cell to ensure immediate response in case of an emergency event.
• Laser scanners: Two laser scanner devices (Figure 10c) have been strategically placed within the robotic cell layout to monitor the presence of human operators and automatically stop the assembly process in case an operator enters the high-payload robot area. The installation of these laser scanners allows the system to reduce the high-payload robot’s speed when obstacles are detected within the warning area. Additionally, three distinct zone configurations have been defined for each laser scanner, with the specific zones adapted to the position of the high-payload robot’s base joint.
• Wireless safe bridge: Although the mobile manipulators include their own security system, it is mandatory to connect the mobile robot with the main safety system of the cell. The connection of mobile manipulators with the main safety PLC is based on wireless safe bridge devices, specifically the Bridge E wireless Ethernet bridge from R3 Solutions [32] shown in Figure 10d. This device ensures ultra-reliable low-latency data transmission, creating a secure channel to send safety signals to the central PLC system.
• Mode selector: As previously mentioned, the connection between the engine and gearbox is based on the manual alignment of the two parts by a human operator using the manual guidance technologies implemented on the high-payload robot. However, this task is a critical assembly step as it provides full control to human operators over the high-payload robot. A mode selector device has been integrated, see Figure 10e, to ensure that the manual guidance mode will only be used by the trained operators. To this end, operators of the assembly line insert a specific configuration key inside the mode selector device to enable the manual guidance of the robot in a safe and controlled way.
• Retroreflective sensors: A critical safety step in the integration of mobile manipulators in the assembly process is the approach of the manipulators to the conveyor, ensuring undesired collisions between the robot and the transporting device. Therefore, retroreflective sensors have been installed on the mobile platforms to safely validate the approach of the mobile robotic resources with the conveyor, as shown in Figure 10f. Additionally, reflective tape has been installed along the conveyor, enabling the activation of the retroreflective sensors when both elements are close enough. Thus, the retroreflective sensors send a safe signal to the mobile manipulator’s safety PLC to adapt the safety zones and enable the manipulator’s movements.

The integration of the previously presented devices and safety policies allows the creation of a collaborative environment compliant with ISO 10218-1, ISO 10218-2, and ISO/TS 15066 [33,34,35], paying particular attention to the specific requirements and features of the presented engine assembly line.

6. Digital Twin

Finally, the proposed robotic cell is completed with a Digital Twin, a digital model representing the status of the complete cell (robots, engine, operator, and security status) in real-time. The implemented Digital Twin module seamlessly interfaces with the different elements of the cell through the Robot Operating System (ROS) to receive real-time data of robots and sensors, crucial for creating an accurate digital replica of the physical workstation. This module captures a diverse set of information derived from ROS, including robots’ joint states, enable/disable signals of grippers’ and conveyor’s actuators, precise cues of the picking and placing of gearbox and motor parts, as well as the result of the object detection and quality inspection modules. Additionally, it incorporates data derived from safety scanner sensors, ensuring the reliable visualization of safety zones’ shape. The digital simulation software Visual Components is included to facilitate the visualization of the data managed by the Digitial Twin module. Figure 11 shows the Digital Twin module alongside the physical world of the developed human-centric robotic cell.

For its implementation, the Digital Twin module leverages UDP clients that act as ROS subscribers for fast data exchange. Within Visual Components software, the module employs UDP servers to receive, process, and dynamically alter the properties of the involved components based on the transmitted data. With the implemented UDP communication, the Digital Twin module establishes a real-time connection between ROS and Visual Components software, ensuring the accurate synchronization of the virtual environment with dynamic events.

7. Implementation

The proposed human-centric robotic solution for engine assembly was implemented in a real industrial scenario. Specifically, the complete robotic system was deployed on the Stellantis plant of Mirafiori in Turin, including all the different robots, devices, and components presented in the previous sections. During six weeks, these modules and systems were combined to complete the pilot and validate the complete approach in a real industrial environment.

The central element of the pilot is the circular conveyor that transports the engine along the three different stations, as shown in Figure 1. The conveyor follows a circular design instead of a standard linear shape to facilitate continuous testing and validation as the engine travels along the different stations endlessly. The conveyor is surrounded by the previously presented assembly stations:

• In the initial collaborative high-payload part assembly station, see Figure 12, an operator assisted by a Comau Aura collaborative industrial robot fixes the gearbox to the engine body using the hand-guiding device, using augmented reality to interact with the complete robotic system.
• After this initial step, the operator fits the engine cover to the main body by placing several bolts manually. Once the engine cover is in position, the complete engine moves along the mobile screw fastening station where an autonomous mobile manipulator fastens these screws as shown in Figure 13.
• Finally, when the engine reaches the robotized quality inspection station, a second mobile manipulator captures various images to perform the quality inspection and verify the previous steps of the assembly. The operator receives real-time feedback on the quality inspection, see Figure 14, as well as corrective measures whenever the quality system detects incorrectly assembled elements.

Additionally, the complete process can be monitored through the Digital Twin of the pilot, verifying the status and position of all robots and assembly stations in real-time even from outside the cell, see Figure 11. This virtual representation offers a valuable tool to cross-check the assembly status, as well as to record real process data that could be further exploited for optimizing different aspects of the assembly stations.

The complete architecture of the deployed system is divided into different hardware elements, as depicted in Figure 15:

• The central element is the central ROS PC, which manages all the robot and safety information and triggers the different robot actions. It also distributes all required information to the digital twin PC and interacts with the operator through the AR PC.
• The Safety PLC gathers all safety information from the safety devices, industrial robot, and mobile manipulators. After processing this safety information, the processed information is sent to the central ROS PC.
• The digital twin PC receives the cell status information from the central ROS PC, storing it and digitally representing the state of the whole cell.
• The AR PC receives the high-level operator interaction commands from the AR hardware, processing the information and sending it to the central ROS PC. Afterwards, the system feedback is presented to the operator through the AR hardware.

The complete pilot has been deployed following the current safety standards (ISO 10218-1, ISO 10218-2, and ISO/TS 15066) as a crucial aspect of implementing a human-centric and collaborative environment for operators. Risk analysis for all three assembly stations has been carried out, focusing on the particularities of each station and the specific interaction between operators and robots in the assembly operations.

8. Validation

The proposed human-centric pilot was validated during the deployment phase on the Stellantis plant of Mirafiori. Due to the large number of elements and robots included in the robotic cell, the validation is divided into assembly stations, trying to assess the specific features of each location. Therefore, the validation measures the stability and success rate of the assembly, as well as the operators’ feedback on the technologies applied for the human-robot interaction in an attempt to evaluate the human-centric aspect of the complete cell.

8.1. Collaborative High-Payload Part Assembly

During the initial phase of the assembly process, operators attach the gearbox to the main body of the engine, making use of hand guidance technology to manipulate the gearbox while AR provides feedback about the process. To assess this human-robot interaction during the assembly process, 37 assembly line operators utilized the system. Each operator was able to test the developed AR application and interact with the high-payload robot of the assembly line through the developed interfaces of the application. Afterwards, the feedback from these operators was collected, focusing on the usability of both technologies and divided into three different questions:

• Added value of AR-based application during the execution of the assembly process: The first question focuses on the added value of the developed AR-based application on this initial station. 33 out of the 37 workers thought it was very profitable for them during the execution of the gearbox assembly process as it offered a detailed visualization of how to perform the required tasks with a user-friendly interface. 3 operators were neutral about the added value of the AR application, while 1 of them categorized the application as a non-added value technology.
• Complexity of AR-based application for assembly line’s operators: The second question focuses on the complexity of the AR application for the operators. 30 out of 37 operators described the AR-based application as "easy to use" during the axle alignment tasks. According to the collected feedback, the assembly instructions for the operators were very clear and presented very precisely to the operators through the virtual world of the AR application. 5 operators were neutral about the complexity of the application. Only 2 operators stated that the developed solution was difficult to use due to the high brightness and the contrast of the objects included in the virtual world of the AR application.
• Functionality of hand guidance technology: During the tests, operators were able to interact with the high-payload robot through its manual guidance functionality without any calibration procedure required. Out of 37 operators who tested the technology, 34 of them were very surprised about the smoothness of the robotic movements and how easy it was for them to align the engine and gearbox parts with minimum forces applied to the manual guidance joystick installed on the robotic gripper. 2 operators were neutral about the functionality while 1 operator described the developed application as difficult to use due to the high sensitivity of the guidance sensor.

Figure 16 summarizes the obtained operator feedback on the three posed questions, obtaining positive comments about the application of these two technologies in heavy part-handling tasks.

8.2. Mobile Screw Fastening

To validate the quality of the screw fastening operation on the engine that was transported in the conveyor belt, the operation was repeated onsite on Stellantis to verify the consistency and extract different metrics. Specifically, the validation experiment was designed with the following features:

• The engine was placed on the conveyor belt, with 4 different bolts accessible for the robot, as shown in Figure 17.
• The engine is transported at a speed of 100 mm/s, starting the movement when the mobile manipulator triggers the “start” signal when the gearbox assembly area is reached.
• Due to the positioning error of the laser-based navigation system of the mobile platform, the initial relative position between the engine and robot has a variability of up to 10 cm and 5º. Therefore, the engine was not always visible to the robot. In these cases, the start signal was triggered and the robot waited until the engine appeared on the field of view of the camera.
• The screw fastening operation was repeated 25 times, including navigation to the initial screwing positions, screw fastening operation, and navigation back to the robot station area. Therefore, 100 screws were fastened in these 25 rounds.

During the validation, two metrics were recorded, the success rate and the screwing time. When the screw was not totally fastened the operation was considered a failure. The obtained results are listed below:

• 90 of the 100 screws were correctly fastened, obtaining a 90% of success rate. In all the failures, the screw was fastened to some degree although the process stopped due to some oscillations between the robot and engine.
• The mean screw fastening time (per screw) was 8.79 s with a standard deviation of 0.43 s. On average, 4.51 s were spent approaching the screws and 4.28 s fastening once the screwdriver was correctly positioned.

8.3. Robotized Quality Inspection

Following the same methodology applied for the evaluation of the first assembly station, the validation of the robotized solution towards the quality inspection of the engine and gearbox assembly collected the feedback received from the same operators who tested the hand guidance technology. These 37 operators also used the AR application to perform the required corrective actions on incorrectly assembled parts. 33 out of 37 operators were very satisfied with the functionality of the AR application during the execution of the corrective actions. Based on their input, the holograms visualized inside the virtual world of the AR application, presenting how the corrective actions should be carried out, were very useful and easy to understand. 3 operators were neutral about the integration of the robotized solutions for the quality inspection process. Finally, 1 operator was unsatisfied with the proposed approach due to the complexity of visualizing the holograms inside the virtual world of the AR application.

Figure 18 summarizes the provided feedback about the use of AR technologies in the assembly error correction.

9. Conclusions and Future Work

The presented paper proposes a human-centric robotic solution for gearbox and engine assembly, emphasizing Industry 5.0 aspects such as human-robot interaction and safety. In this sense, the described solution poses a design where operators are the central element of the cell, being able to interact with the robotic system using wearable devices and augmented reality while industrial robots and autonomous mobile manipulators operate around. From the robot-assisted gearbox assembly to the autonomous screw fastening and quality inspection, several engine assembly sub-steps are considered in the definition of this human-centric assembly line.

The proposed human-centric robotic design has been deployed and validated at the Mirafiori plant of Stellantis, integrating the different elements and subsystems in a complete and functional pilot following the current safety standards. The deployed pilot has been validated from different perspectives, including tests with 37 assembly line operators. The results demonstrate the suitability of the approach, obtaining promising results in all the stations of the presented assembly line.

In future steps, several aspects of the pilot could be improved to increase productivity and reliability. The addition of mobile manipulators provides flexibility to robotized assembly lines, allowing the possibility to work on different stations of the line and even working on the parts while they are transported between stations. Even so, the uncertainties introduced by the localization and navigation systems make it necessary to enhance the manipulation and control frameworks to overcome these imprecisions. As an example, the screw fastening control algorithm should be improved to allow a markerless approach based on high-frequency AI-based vision algorithms, increasing the success rate above the achieved 90% to meet the high-quality standards of the automotive industry. Besides, the inclusion of additional human-robot interaction methods should be studied, as AR technologies are still perceived as intrusive by a certain percentage of assembly line operators. The capability to include multi-modal interaction methods would increase the human-centric aspect, increasing the acceptance of this kind of approach.

Finally, due to the large amount of elements and technologies available in the deployed robotic system, the carried-out validation process only tackles some specific aspects of the complete system. As a future step, it is necessary to complete this evaluation with important aspects, such as the practical usability of AR technology in the assembly line as a way to consolidate the technology in the industry. Additionally, a transversal evaluation of the system performance, including task duration, success rate, or error reduction, would highlight the benefits of the implementation of assembly lines based on the Industry 5.0 paradigm.