Design, Implementation and Testing of an EtherCAT-based Network for Multi-modal Distributed Sensing Architectures

Design, Implementation and Testing of an EtherCAT-based Network for Multi-modal Distributed Sensing Architectures Francesco Giovinazzo1,*, Alessandro Perri1, Marco Staiano1, Francesco Grella1, Marco Sartore2, Manuela Adami2, Riccardo Galletti2and Giorgio Cannata1 Abstract— This paper introduces an EtherCAT-based communication infrastructure designed for large-area multi-modal sensor networks in robotic applications. The system addresses critical application-oriented challenges, such as bandwidth optimization, cable routing efficiency, network scalability and fault tolerance. The proposed architecture is experimentally validated with ProxySKIN, a distributed sensing technology integrating arrays of tactile and proximity sensors, designed to cover large areas of robotic manipulators. The implemented solution leverages a redundant EtherCAT network driven by custom-made miniaturized slave devices to collect comprehensive data regarding the robot’s surrounding space. Bench-test experiments conducted on over 6400 pressure sensors and 20 multi-zone proximity sensors show reliable data collection with minimal communication latency through a unique EtherCAT backbone. I. INTRODUCTION The removal of physical barriers between humans and robots in industrial settings has paved the way for innovative manufacturing models, driving the transition toward collaborative robotics. Industrial robots must operate and interact safely with the surrounding space without relying on environment-dependent sensing infrastructures (e.g. cameras, motion capture systems) in order to increase the flexibility of the shop-floor configuration and reduce setup time and costs. To enhance the autonomy, perception, and cognitive capabilities of robots, the integration of networks of heterogeneous sensors across large areas of the robotic platform represents a promising solution. Existing literature offers numerous solutions for distributed multi-modal sensing architectures, such as [1], [2] and [3] and [4], relying on different sensor types to provide a comprehensive representation of the robot surrounding space. The effectiveness of these technologies is demonstrated in a wide range of applications: In [5] and [6] authors present a novel approach to realize whole-body tactile interactions with a humanoid robot equipped with HEX-O-SKIN, a selforganizing, multi-modal artificial skin. In [7] authors demonstrate how multi-modal sensor arrays replacing the spacer disks of a continuum robot can scan the robot’s peripersonal space and detect human contact for safe Human-Robot Interaction. In [8], the technology demonstration implemented 1Department of Informatics, Bioengineering, Robotics and Systems Engineering (DIBRIS), Universit` a di Genova, Via all’Opera Pia 13, 16145 Genova, Italy 2ElbaTech Srl, Via Roma 10, 57030 Marciana, Italy ∗Corresponding author e-mail: [email protected] This work was supported by the SESTOSENSO (HORIZON EUROPE Research and Innovation Actions under GA number 101070310). Fig. 1. Schematic representation of the EtherCAT-based communication architecture for distributed sensing networks. Five main components are highlighted: the EtherCAT master, the physical bus connecting the Host PC to the EtherCAT slaves, the network of slave devices, the Intelligent Hub Board (IHB) microcontroller and patches of different types of sensors. for the European project CoLLaboratE [9] showcases the safe and effective utilization of a heavy-duty industrial robot in physical contact with a human operator. The collaborative assembly and inspection task is made possible by integrating four handles covered with distributed tactile sensors onto the robot gripper, used by the operator to convey his intention to interact. In these applications, to ensure a reliable and efficient data communication between distributed sensing elements and the central processing unit, it’s essential to design suitable network infrastructures considering performance requirements such as bandwidth, latency, determinism and robustness. The literature has many contributions on real-time communication within the context of distributed control. For instance, authors in [10] present and evaluate the performance of an EtherCAT-based communication network for interconnecting servo motors and a large number of sensors in real-time motion control systems. In [11] authors verify the timing accuracy of an EtherCAT-based real-time motion control system to actuate a differential drive mobile robot. In this paper, we introduce an EtherCAT-based communication network designed for distributed multi-modal sensing architectures. The proposed solution addresses typical infrastructure challenges, such as bandwidth optimization, cable routing efficiency, scalability and fault tolerance, while simplifying the interconnection and integration of distributed sensors over large area surfaces. The implemented solution is tested with ProxySKIN [12], a large-area sensing network combining proximity and tactile sensors. Benchmark tests are 2024 IEEE 20th International Conference on Automation Science and Engineering (CASE) August 28 - September 1, 2024. Bari, Italy 979-8-3503-5851-3/24/$31.00 ©2024 IEEE 830 2024 IEEE 20th International Conference on Automation Science and Engineering (CASE) | 979-8-3503-5851-3/24/$31.00 ©2024 IEEE | DOI: 10.1109/CASE59546.2024.10711599 Authorized licensed use limited to: Centre for Research and Technology (C.E.R.T.H.). Downloaded on September 15,2025 at 13:36:42 UTC from IEEE Xplore. Restrictions apply. conducted on over 6400 tactile elements and 20 multi-zone Time-of-Flight (ToF) sensors, capable of comprehensively covering one link of an industrial manipulator. Building upon previous research outlined in [13] and [14], the main contributions of this article are: •the performance evaluation of the EtherCAT communication infrastructure applied to a distributed multi-modal sensing architecture; •the design and hardware implementation of custommade miniaturized EtherCAT slave boards; •an algorithm designed to efficiently manage and transmit large amount of sensor data throughout the network. The article is organized as follows: Section II introduces the reference multi-modal sensing architecture and provides a system-level description of the network. It also covers the development of the customized EtherCAT slave devices. Section III presents the EtherCAT performance metrics analysed and the solution implemented to efficiently acquire a large amount of data. Section IV discusses relevant performance tests and provides preliminary qualitative results on the acquired data. Conclusions follow. II. NETWORK ARCHITECTURE Designing a large-area robot skin requires consideration of several aspects that significantly impact the efficiency and flexibility of the overall architecture. Apart from the hardware design, manufacturing and deployment of the sensing technology, particular attention must be paid to the communication infrastructure supporting the network of distributed sensors. Crucial aspects such as network modularity, scalability, real-time performance and bandwidth optimization must be thoroughly addressed. Additionally, a streamlined physical communication interface should be preferred to simplify the integration process along the robot multi-body chain. To this aim, we have developed an embedded communication infrastructure based on EtherCAT specifically designed for large-area sensor networks, satisfying most of the previous requirements. From a system level perspective, the architecture features five main components, schematically represented in Figure 1: •Groups of distributed interconnected sensors, called ’patches’, gathering heterogeneous data about the robot surrounding space; •The Intelligent Hub Board (IHB) [15], a microcontroller integrating an ARM Cortex-M3, programmed to locally acquire data from a patch of sensors through four dedicated SPI buses and to perform preliminary filtering and processing operations on the raw data. •The EtherCAT slave board, featuring an ARM CortexM4 XMC4300 microcontroller [16] and two EtherCAT PHY components, driving the EtherCAT communication at a hardware level. The device retrieves data from an associated IHB through an SPI interface and manages the underlying EtherCAT communication protocol, ensuring high performance and real-time capabilities. Fig. 2. Sample of ProxySKIN (52 cm x 78 cm), the distributed multi-modal sensing architecture used to test the implemented EtherCAT communication infrastructure. This sample integrates two arrays with 10 proximity sensors and twelve patches of CySKIN with 515 pressure sensors. •The EtherCAT physical communication bus, interconnecting multiple ETherCAT slaves with a centralized master node through a single or multiple Ethernet communication backbones. •The EtherCAT master node, acquiring and processing low level data from the distributed sensing network for high level applications. This component is based on the Simple Open EtherCAT Master Library (SOEM), implemented on a standard PC with a real-time patched Linux kernel. The proposed communication infrastructure is evaluated with patches of proximity and tactile sensors. However, its modular and layered structure enables it to operate independently from the connected sensing devices, thereby enhancing the flexibility of the overall architecture. Moreover, the network 831 Authorized licensed use limited to: Centre for Research and Technology (C.E.R.T.H.). Downloaded on September 15,2025 at 13:36:42 UTC from IEEE Xplore. Restrictions apply. Fig. 3. EtherCAT slave PCB measuring 4.81 x 3.30 cm. a) Bottom layer: in yellow are highlighted the mezzanine connectors to the IHB microcontroller, that is ”piggy-backed” to the EtherCAT PCB. Typical RJ-45 connectors are replaced with 2 small SMD connectors, positioned at the top of the board. Bypass capacitors and other decoupling parts are positioned very close to the microcontroller power pins. b) Top layer: The PHY-drivers labeled “IN” and “OUT”, highlighted in green, are not coupled to isolation transformers. The XMC4300 controller is below them, 45 degrees-oriented, to optimize the trace routing. topology can be easily reconfigured, and the size can be scaled up by connecting additional slave boards, along with their corresponding sensing units. The integration of the large-area sensing architecture on the robot body is simplified due to the reduced number of physical communication buses interconnecting the EtherCAT slave devices to the master node. A. ProxySKIN ProxySKIN is the reference multi-modal sensing architecture used to test the EtherCAT-based communication infrastructure. This technology, envisioned by the European project Sestosenso [17], integrates networks of discrete miniaturized proximity sensors and tactile sensors, meant to cover large areas such as the links of a robot. ProxySKIN is a skin-like sensory system that provides comprehensive and continuous proxy-tactile perception of the robot surrounding space. As thoroughly described in [12], it relies on patches of CySkin to collect tactile information [4] and arrays of STM VL53L8CX ToF sensors [18] to acquire knowledge about the environment’s geometry. CySkin has a modular structure based on flexible triangular elements that host up to 11 pressure sensors. The ToF sensors, instead, feature a multi-zone distance measurement capability with 64 separate zones, a Field of View of 45◦and a range up to 400 cm. A sample of ProxySKIN is shown in Figure 2. B. Custom EtherCAT slave PCB design To drive the EtherCAT communication and acquire data from the sensing devices, we designed a custom EtherCAT slave PCB, based on the XMC4300 Infineon microcontroller [16]. This high-performance and energy-efficient integrated circuit (IC) was chosen among multiple potential candidates primarily due to its built-in EtherCAT Slave interface, capable of 100 Mbit/s transfer rates, with 2 MII (Media Independent Interface) ports ready to form the typical input/output pair of an EtherCAT unit. From an integration perspective, our primary objective was to design a miniaturized circuit board that could be easily embedded onto the links of a robot manipulator. In our application the distributed sensing network is confined to the robot body surface and we demonstrated that this setup is comparable to a backplane/mid-plane oriented EtherCAT application, where communication can be achieved without using isolation transformers. We verified that the PHY-toPHY connection could be implemented using simple ceramic capacitors, thereby eliminating bulky transformers and RJ-45 connectors. A valuable condition to prevent signal noise issues is the presence of a single supply line, the +3.3V vs GND. This simplifies the power connections on the board via two dedicated power planes in a 4-layers PCB stack. With the outer two layers used for interconnections between the surface mount components, the inner power and ground plane are wide and cover the whole area without isolated ground ”islands”. Indeed, IC ground pins located in a ground island have no current return path to the ground plane and may induce unpredictable behavior of the related integrated circuit. The power supply is decoupled to the low-impedance ground plane with a high quality electrolytic capacitor at the input port. This minimizes low frequency noise on the supply lines. At each individual stage, local high-frequency filtering is required at the individual IC package power pins, located as close as possible to the respective pins. The parts positioning then plays an important role in the realization of an optimized board. Our strategy is to place the bypass power capacitors (and other power-related parts) on the bottom layer in close proximity to the power pins. The integrated circuits lays onto the top layer, as shown in Figure 3. III. ETHERCAT AND SENSORS DATA MANAGEMENT Building on prior research regarding distributed tactile sensors [13], [14], we chose the EtherCAT communication standard to manage the low-level networking infrastructure for ProxySKIN. EtherCAT, standardized in IEC 61158, boasts a 100 Mbps data rate, making it the fastest industrial Ethernet technology suitable for real-time performance requirements. This protocol overcomes limitations of other Ethernet solutions by employing a principle known as ”processing on the fly”. The master node controls data exchange over the network, initiating communication with the slave devices. Instead of sending one data packet per slave per cycle per direction, the master node sends one or more frames through all network nodes, which extract and insert their own data onto a dedicated section of the frame without blocking the datastream. Besides optimal network performance, EtherCAT features a scalable architecture and 832 Authorized licensed use limited to: Centre for Research and Technology (C.E.R.T.H.). Downloaded on September 15,2025 at 13:36:42 UTC from IEEE Xplore. Restrictions apply. TABLE I ROUND TRIP TIME ANALYSIS (Tdelay )DEPENDING ON THE NUMBER OF ETHERCAT SLAVES (nslaves )AND TOFSENSORS (nToFs)CONNECTED TO TO EACH SLAVE DEVICE.nf rames REPRESENTS THE NUMBER OF ETHERNET FRAMES NEEDED TO TRANSMIT THE PAYLOAD OF ALL THE SLAVE DEVICES WITHIN AN ETHERCAT CYCLE TIME.pslave REPRESENTS THE SENSOR DATA PAYLOAD OF EACH SLAVE DEVICE EXPRESSED IN BYTES nslaves nToFs 1 2 3 4 Tdelay [µs]nf rames Tdelay [µs]nf rames Tdelay [µs]nf rames Tdelay [µs]nf rames 1 31.755 1 57.515 1 83.275 1 109.04 1 2 60.470 1 111.99 1 166.55 2 218.07 2 3 89.185 1 169.51 2 246.79 2 327.11 3 4 117.90 1 223.98 2 330.06 3 436.14 4 5 149.66 2 281.50 3 413.34 4 545.18 5 6 178.37 2 335.97 3 493.57 4 654.21 6 7 207.09 2 393.49 4 576.85 5 763.25 7 8 235.64 2 447.80 4 659.96 6 869.08 7 9 264.36 2 502.28 4 740.20 6 978.12 8 10 296.11 3 559.79 5 823.47 7 1087.2 9 11 324.82 3 614.27 5 906.75 8 1196.2 10 12 353.54 3 671.78 6 986.98 8 1305.2 11 pslave [bytes] 324 646 968 1290 flexible topology, enabling the interconnection of up to 216 slave devices in various configurations. These characteristics also improve cable routing efficiency, as the physical network infrastructure can be streamlined to a single communication backbone passing through all slave devices up to the master node. Finally, EtherCAT’s diagnostic capability ensures reliable detection of bit faults through the evaluation of the CRC checksum. The aforementioned specifications make the EtherCAT standard an optimal solution for managing communication in large-area sensing architectures. However, the EtherCAT protocol must be adapted to comply with the requirements of our application, such as minimizing the number of slave devices in the network, while acquiring data from a large number of sensors and optimizing the bandwidth consumption. Slave nodes are typically associated with sensors, drivers, or actuators that provide small payloads (tens of bytes) at a high rate. Just a few concatenated Ethernet frames, with a maximum capacity of 1500 bytes, are sufficient to transfer data from numerous slave devices to the master node. In our context, we aim to use a small number of EtherCAT slaves to gather a large amount of data from patches of sensors operating at low frequencies. For instance, an array of 10 Time-of-Flight sensors in ProxySKIN provides 3220 bytes every 66.7 ms, while 100 CySkin modules (approximately 1100 tactile elements) must transmit 2200 bytes at 20 Hz to the master node. A single Ethernet frame cannot accommodate the entire data payload provided to the slave device by a patch of sensors, but the collected information can be divided into multiple ’segments’ fitting inside the data packets, that are sequentially forwarded to the master node in consecutive cycles. In ProxySKIN, data collected by 10 ToF sensors by a single EtherCAT slave can be split into three segments of 1074 bytes, which are sent to the master application in three subsequent EtherCAT cycles. An alternative solution would be to limit the number of sensors connected to a slave node to fit the data payload into a single Ethernet frame. However, this approach would require more IHB microcontrollers, slave devices, power supplies, and networking cables, thereby increasing the complexity and cost of the sensing infrastructure. To evaluate EtherCAT network performance, we considered six interrelated metrics, that directly impact the robot’s response time and behavior: the round-trip time, the slave minimum cycle time, the master minimum cycle time, the EtherCAT cycle time and the overall data transmission period. All these quantities are influenced by the number of sensors handled by the network and their data payload. 1) The round-trip time: is defined as the time an EtherCAT frame takes to start from the host PC and return to the host after traveling through all the slave device nodes. This parameter, also known as frame latency, only depends on the hardware propagation delay times and affects the master minimum cycle time. It can be computed as explained in [13] and [14] as: Tecat =Nsl ×(Tecat f wd +Tdata) +Ncables ×Tmedium +nf r ×Teth +Tecat ov +Tpad (1) where: Nsl is the number of slaves; Tecat f wd is the time needed for each slave to forward the data frame to the next node; Tdata is the time required to send the specific payload of each slave node; Ncables is the number of network cables; Tmedium is the delay introduced by the signal propagation; nf r is the number of Ethernet frames that pass through the network; Teth is the Ethernet overhead time; Tecat ov is the EtherCAT overhead time; Tpad is the time needed to send additional padding bytes of the Ethernet frame. 2) The slave minimum cycle time: represents the minimum time required by an EtherCAT slave to perform cyclical sequences of calculations and data copying tasks [19]. This software-dependent parameter does not affect the round-trip time, as the processing operations run in parallel to the physical propagation of the frames throughout the network, without blocking the telegram datastream. However, 833 Authorized licensed use limited to: Centre for Research and Technology (C.E.R.T.H.). Downloaded on September 15,2025 at 13:36:42 UTC from IEEE Xplore. Restrictions apply. TABLE II SLAVES MINIMUM CYCLE TIME STATISTICAL ANALYSIS (Tmin cycle ). THE ANALYSIS CONSIDERS 20000 SAMPLES COLLECTED BY A VARYING NUMBER OF ETHERCAT SLAVES (nslaves ). slave id REFERS TO THE UNIQUE IDENTIFIER ASSIGNED TO THE SLAVE DEVICES. MEAN VALUES (µ)AND STANDARD DEVIATIONS (σ)ARE EXPRESSED IN MICROSECONDS [µs]. nslaves slave id 12345 µ[µs]σ[µs]µ[µs]σ[µs]µ[µs]σ[µs]µ[µs]σ[µs]µ[µs]σ[µs] 1 465.59 0.04 / / / / / / / / 2 540.89 1.93 465.59 0.02 / / / / / / 3 624.61 2.62 603.66 2.27 465.84 0.04 / / / / 4 707.87 3.56 687.85 3.91 672.79 0.02 466.12 0.01 / / 5 780.83 5.68 735.02 5.08 670.96 1.98 715.47 3.38 466.19 0.05 it affects the computation of the master minimum cycle time. The software operations that impact on the minimum cycle time include: copying the output process data from the SyncManager into the local memories, processing the output data, reading the input signal from the local memory, processing the input data, and copying the input process data into the input data area of the SyncManger. 3) The master minimum cycle time: is defined as the shortest time interval the EtherCAT master needs to cyclically perform two consecutive input/output data exchanges with all network devices. This parameter depends on both the round trip time (hardware contribution) and the slave minimum cycle time (software contribution). 4) The EtherCAT cycle time: represents the time interval between the transmission of two subsequent datagrams throughout the network. 5) The overall data transmission period: represents the time required to transmit the full sensors data payload from the slave devices to the host PC by means of consecutive input/output data exchanges. This parameter is a multiple of the EtherCAT cycle time and depends on the number of ’segments’ dividing the collected sensors data. IV. PERFORMANCE TESTS AND RESULTS The performance tests carried out for the implemented EtherCAT communication infrastructure are essential to understand the data propagation delays to the host PC and validate the use of the EtherCAT protocol for large-area sensor networks in robotics applications. The analysed parameters include the round-trip time, the slaves minimum cycle time, the master minimum cycle time, the EtherCAT cycle time and the overall data transmission period. The experimental setup used to validate the EtherCAT communication framework consists of three main components: (a) a sample of ProxySKIN, shown in Figure 2, characterized by 12 patches of 515 tactile sensors, and 2 arrays of 10 proximity sensors, acquiring data at 15 Hz; (b) 5 slave devices, equipped with an XMC4300 Infineon microcontroller, configured in a non-redundant, daisy-chain topology. Two slave devices are connected to the arrays of ToF sensors and three slaves acquire data from patches of pressure sensors; (c) A master node implemented with the SOEM library on a standard PC with a real-time patched Linux kernel. A. Theoretical round-trip time evaluation In this test, we analyse the hardware propagation delay times of data frames from the master node to the slave devices in a single EtherCAT cycle time. To this end, we consider slave devices with a reduced sensor data payload, reaching at most 1480 bytes, which corresponds to the maximum load capacity of an Ethernet frame. Higher data payloads are not considered because they would be divided into multiple ’segments’ sent to the master in subsequent cycles, as explained in Section III. Table I presents the results obtained when applying the mathematical model described in Equation (1) to a varying number of EtherCAT slaves (ranging from 1 to 12) and ToF sensors (ranging from 1 to 4). Results show that the frame delay affecting the master minimum cycle time is limited to hundreds of microseconds and in the worst case configuration it never exceeds 1.5 ms. These values are suitable for our application, since the sensors’ sampling rate are rather slow (15 Hz). The same considerations apply to CySkin and other sensing devices with comparable performance. B. Slaves minimum cycle time The slaves minimum cycle time, required to perform cyclical sequences of calculations and data copying, is statistically analysed by reading the standard objects 0x1C32 and 0x1C33, storing the SyncManager input/output parameters [20]. In particular, the entry 0x1C3x:5 provides the minimum cycle time (expressed in nanoseconds) supported by the slave. Table II presents the results obtained for different network configurations, by analysing 2 ×104data samples from a varying number of EtherCAT slaves (nslaves) ranging from 1 to 5. By increasing the number of EtherCAT slaves, and consequently the number of sensors connected to the network, we observe higher mean values and standard deviations for the slaves’ minimum cycle time. The increased number of datagrams circulating through the network requires slave devices to perform additional software operations on the input/output process data objects. C. Master minimum cycle time To evaluate the master minimum cycle time, we statistically analyze the actual time latency between the transmission and reception of concatenated Ethernet frames on 834 Authorized licensed use limited to: Centre for Research and Technology (C.E.R.T.H.). Downloaded on September 15,2025 at 13:36:42 UTC from IEEE Xplore. Restrictions apply. TABLE III STATISTICAL ANALYSIS ON THE FRAME DELAYS (Tdelay), ETHERCAT CYCLE TIME (Tecat )AND OVERALL DATA TRANSMISSION PERIOD (Tf ull data). THE ANALYSIS CONSIDERS 30000 SAMPLES COLLECTED BY A VARYING NUMBER OF ETHERCAT SLAVES (nslaves). MEAN VALUES (µ)AND STANDARD DEVIATIONS (σ)ARE EXPRESSED IN MICROSECONDS [µs]. nslaves 12345 µ[µs]σ[µs]µ[µs]σ[µs]µ[µs]σ[µs]µ[µs]σ[µs]µ[µs]σ[µs] Tdelay 524.33 60.79 524.33 62.31 744.61 65.98 753.12 66.76 991.94 73.02 Tecat 782.06 70.29 831.06 72.65 944.46 71.87 950.95 77.11 1.14x10382.84 Tf ull data 2.34x103210.86 2.49x103217.95 2.83x103215.63 2.85x103231.33 3.41x103248.53 Fig. 4. Screen capture of the oscilloscope showing important performance parameters of the EtherCAT-based communication network. Reported data refer to an EtherCAT network with 2 slave devices collecting data from patches of pressure sensors. Yellow signal is triggered by the thread of the master EtherCAT right before a transmission of the data frames to the slave devices and immediately after the reception of the data frames from the network. The time interval between two consecutive Ethernet frame transmissions represent the EtherCAT cycle time. Cyan signal: is probed on a pin of the input PHY driver of the first EtherCAT slave and represents the data transmission over the MII interfce to the XMC4300 microcontroller. Magenta signal represents the minimum cycle time of the first slave, i.e. the time required to perform cyclical sequences of calculations and data copying tasks. Green signal represents the minimum cycle time of the second slave. the master node. In Table III, Tdelay shows the frame delay observed for our experimental setup in different network configurations, by varying the number of EtherCAT slaves (nslaves) from 1 to 5. The statistical analysis focuses on 3×104samples and shows that the master node can perform an input/output data exchange with the slave devices within hundreds of microseconds. The resulting values are significantly higher than the theoretical results computed for the round-trip time, because we also consider preliminary processing operations on the raw data. However, the master minimum cycle time depends also on the slaves minimum cycle time. Indeed, software operations performed by the slave devices on the process data run in parallel to the physical propagation of the EtherCAT datagrams through the network. Therefore, when the cyclical Ethernet frames return to the master node, EtherCAT slaves might still be processing input/output data. This behaviour can be clearly seen in Figure 4, presenting qualitative results on a network with two slave devices. On the screen capture of the oscilloscope we can notice 4 signals: •Yellow signal: refers to the EtherCAT master thread running on the PC. The two spikes highlighted in red indicate a new transmission of Ethernet frames to the slave devices. The spike highlighted by the green arrow indicates the reception of the Ethernet frames belonging to the previous transmission, coming back from the network. •Cyan signal: indicates the data transmission over the MII interface from the input PHY driver of the first EtherCAT slave to the XMC4300 microcontroller. The time interval between the first yellow spike and the data transmission on the MII interface represents the 835 Authorized licensed use limited to: Centre for Research and Technology (C.E.R.T.H.). Downloaded on September 15,2025 at 13:36:42 UTC from IEEE Xplore. Restrictions apply. Fig. 5. Picture of the experimental setup and corresponding visualization of data collected by ProxySKIN. Data acquired through the EtherCAT network are processed and displayed via the ROS framework. The red dots represent the active pressure sensors, while the proximity point cloud is visualized in white. hardware propagation delay of the Ethernet frame from the master node, consistent with the theoretical values shown in section IV-A. •Magenta signal: refers to the minimum cycle time of the second slave device. •Green signal: refers to the software of the first EtherCAT slave. Specifically it represents the slave minimum cycle time, triggered by the SyncManager event and kept active until the end of the processing operations on the input/output process data objects. When the master node receives the Ethernet frames (second spike of the yellow signal) both EtherCAT slaves are still engaged in copying and processing input/output data (green and magenta signals). Therefore, to ensure a reliable and effective data communication throughout the network we must guarantee that the EtherCAT slaves have updated their process data objects before transmitting a new datagram. The master minimum cycle time is qualitatively evaluated by analyzing the signals on the oscilloscope for different network configurations. In particular, we compute the time interval between the transmission of the Ethernet frames from the master node and the end of processing operations on the input/output process data. The master minimum cycle times computed for an increasing number of EtherCAT slaves are respectively: 700 µswith a single slave, 780 µswith 2 slaves , 850 µswith 3 slaves, 910 µswith 4 slaves and 103µswith 5 slaves. D. EtherCAT cycle time and data transmission period To compute the EtherCAT cycle time, we consider a constant delay in addition to the frame delays analyzed in section IV-C. Hence, we ensure that the EtherCAT slaves have updated their process data objects before the transmission of a new sequence of Ethernet frames. Experimental results show that a delay of about 250 µsis acceptable to ensure a reliable data communication with the EtherCAT slaves. In Table III, Tecat represents the EtherCAT cycle time computed for different network configurations, by varying the number of slave nodes (nslaves) from 1 to 5. The statistical analysis focuses on 3 ×104samples and shows that the master node can perform two consecutive input/output data exchanges with the slave devices within few milliseconds. The overall data transmission period for our distributed sensing architecture is computed by concatenating several EtherCAT cycles as explained in section III. In Table III, Tf ull data shows the data transmission period computed for different network configurations, by varying the number of slave nodes (nslaves) from 1 to 5. The statistical analysis focuses on 3 ×104samples and shows that the master node can successfully transfer the overall sensors data payload within few milliseconds. Results show that the data transmission period has little relevance when compared to the sampling interval of the considered sensors (66.7 ms), allowing the master to collect the full data payload before the next sampling cycle. Figure 5 shows the visualization of the ProxySKIN sensing architecture collected by the EtherCAT network during a physical interaction experiment. The full video of the interaction demo can be found at this link. V. CONCLUSIONS In this paper we presented an EtherCAT-based communication network designed for multi-modal distributed sensing architectures. We introduced the problem of data acquisition from large-area sensor networks, analyzing performance requirements for a reliable and efficient data exchange. We showed that an EtherCAT-based communication infrastructure is beneficial for the integration of large-area sensors on robotic manipulators and can be used for applications with real-time performance requirements. We presented an algorithm that can manage and transmit large amount of 836 Authorized licensed use limited to: Centre for Research and Technology (C.E.R.T.H.). Downloaded on September 15,2025 at 13:36:42 UTC from IEEE Xplore. Restrictions apply. sensor data throughout the network. Moreover, we designed custom-made EtherCAT slave PCBs that minimize space occupancy, which is a fundamental requirement for integrating sensors over large areas. Finally, we validated the overall architecture with the ProxySKIN technology, a distributed network integrating tactile and proximity sensing elements. The analysis of performance metrics such as the round-trip time, the slaves minimum cycle time, the EtherCAT cycle time and the overall data transmission period demonstrated that a large amount of sensor data can be reliably sent to the host PC within few milliseconds. The next steps of our research will include the assessment of the EtherCAT-based communication infrastructure with a larger sensor network integrated on the links of a UR10e robotic manipulator. REFERENCES [1] P. Mittendorfer and G. Cheng, “Humanoid Multimodal Tactile-Sensing Modules,” IEEE Transactions on Robotics, vol. 27, no. 3, pp. 401–410, jun 2011. [2] C. Abah, A. L. Orekhov, G. L. Johnston, P. Yin, H. Choset, and N. Simaan, “A Multi-modal Sensor Array for Safe Human-Robot Interaction and Mapping,” in 2019 International Conference on Robotics and Automation (ICRA). Montreal, QC, Canada: IEEE, may 2019, pp. 3768–3774. [3] Y. Zhou, J. Zhao, P. Lu, Z. Wang, and B. He, “Tacsuit: A wearable large-area, bioinspired multimodal tactile skin for collaborative robots,” IEEE Transactions on Industrial Electronics, vol. 71, no. 2, pp. 1708–1717, 2024. [4] A. Albini, S. Denei, and G. Cannata, “Enabling natural human-robot physical interaction using a robotic skin feedback and a prioritized tasks robot control architecture,” in 2017 IEEE-RAS 17th International Conference on Humanoid Robotics (Humanoids). Birmingham: IEEE, nov 2017, pp. 99–106. [5] P. Mittendorfer, E. Yoshida, T. Moulard, and G. Cheng, “A general tactile approach for grasping unknown objects with a humanoid robot,” in 2013 IEEE/RSJ International Conference on Intelligent Robots and Systems, 2013, pp. 4747–4752. [6] E. Y. P. Mittendorfer and G. Cheng, “Realizing whole-body tactile interactions with a self-organizing, multi-modal artificial skin on a humanoid robot,” Advanced Robotics, vol. 29, no. 1, pp. 51–67, 2015. [Online]. Available: https://doi.org/10.1080/01691864.2014.952493 [7] C. Abah, A. L. Orekhov, G. L. H. Johnston, and N. Simaan, “A multimodal sensor array for human–robot interaction and confined spaces exploration using continuum robots,” IEEE Sensors Journal, vol. 22, no. 4, pp. 3585–3594, 2022. [8] A. Sidiropoulos, F. Dimeas, D. Papageorgiou, T. Prapavesis Semetzidis, Z. Doulgeri, A. Zanella, F. Grella, K. Sagar, M. Jilich, A. Albini, G. Cannata, and M. Zoppi, “Safe and effective collaboration with a high-payload robot: A framework integrating novel hardware and software modules,” IEEE Robotics Automation Magazine, pp. 2–11, 2023. [9] “Collaborate project,” 2018, https://collaborate-project.eu/. [10] L. Wang, M. Guoli, and J. Wang, “Implementing for Networked Motion Control System with Large-Capacity Data Acquisition,” Procedia Engineering, vol. 15, pp. 28–32, Jan. 2011. [11] R. Delgado, S.-Y. Kim, B.-J. You, and B.-W. Choi, “An EtherCATbased real-time motion control system in mobile robot application,” in 2016 13th International Conference on Ubiquitous Robots and Ambient Intelligence (URAI), Aug. 2016, pp. 710–715. [12] F. Giovinazzo, F. Grella, M. Sartore, M. Adami, R. Galletti, and G. Cannata, “From cyskin to proxyskin: Design, implementation and testing of a multi-modal robotic skin for humanndash;robot interaction,” Sensors, vol. 24, no. 4, 2024. [Online]. Available: https://www.mdpi.com/1424-8220/24/4/1334 [13] E. Baglini, G. Cannata, and F. Mastrogiovanni, “Design of an embedded networking infrastructure for whole-body tactile sensing in humanoid robots,” in 2010 10th IEEE-RAS International Conference on Humanoid Robots, 2010, pp. 671–676. [14] E. Baglini, S. Youssefi, F. Mastrogiovanni, and G. Cannata, “A realtime distributed architecture for large-scale tactile sensing,” in IEEE International Conference on Intelligent Robots and Systems, Sep. 2014. [15] “Cyskin project,” 2015, https://www.cyskin.com/. [16] “Datasheet xmc4300 microcontroller,” 2023, https://www.infineon. com/dgdl/Infineon-XMC4300-DataSheet-v01 02-EN.pdf?fileId= 5546d462525dbac40152abc34fde16d8. [17] “Sestosenso project,” 2022, https://sestosenso.eu/. [18] “Datasheet vl53l8cx - low-power high-performance 8x8 multizone time-of-flight sensor (tof),” 2023, https://www.st.com/resource/en/ datasheet/vl53l8cx.pdf. [19] “Synchronization modes of an ethercat slave,” 2023, https://infosys.beckhoff.com/english.php?content=../content/1033/ ethercatsystem/2469122443.html&id=. [20] “Standard objects - index 1c32 sm output parameter,” 2023, https://infosys.beckhoff.com/english.php?content=../content/1033/ ep4174/3386579467.html&id=. 837 Authorized licensed use limited to: Centre for Research and Technology (C.E.R.T.H.). Downloaded on September 15,2025 at 13:36:42 UTC from IEEE Xplore. Restrictions apply.