A Soft-Pneumatic Actuator Array for Tactile Stimulation in Preterm Infants

Abstract

Preterm infants in neonatal intensive care units (NICUs) experience impaired neurodevelopment and dysregulated stress responses, partly due to a lack of tactile stimulation. Although massage therapy offers proven therapeutic benefits by stimulating C-tactile afferents through (gentle) dynamic touch, existing methods are limited by clinical staff variability and resource constraints. This work presents a compact soft-pneumatic actuator array (SPAA) utilizing four nylon–TPU actuators (modules) connected in series or in parallel to perform a sequential actuation; this array is designed to deliver safe, shear-free, and massage-like normal compression tailored for preterm infants. Actuator performance was characterized using a load-cell and a pressure sensor under different preloads (10–30 g), establishing operating internal pressures of 20–50 kPa, which produced target force ranges between 0.1 and 0.3 N. Two SPAA architectures were evaluated: (i) parallel manifold with branch resistances and (ii) series chain with graded outlet resistances, using passive fluidic sequencing for controlled activation. The series configuration achieved repeatable sequential actuation with programmable delays, essential for mimicking therapeutic massage patterns. These results demonstrate that passive soft-pneumatic sequencing can reliably deliver dynamic tactile stimuli within neurophysiological and safety constraints, laying the groundwork for standardized, automated neonatal massage therapy in NICUs.

1. Introduction

Preterm birth represents a major global health challenge, with more than 13 million infants born each year prematurely, accounting for over 10% of all live births worldwide [1]. Prematurity interrupts intrauterine development and exposes neonates to the stressful environment of neonatal intensive care units (NICUs), characterized by high noise and light levels, frequent handling, invasive procedures, and prolonged parental separation. These adverse conditions, together with the absence of tactile stimuli such as dynamic human touch, are associated with impaired neurodevelopment, dysregulated stress responses, and long-term disabilities [2,3].

In this context, massage therapy has emerged as an evidence-based intervention to counteract these risks. Through patterned tactile stimulation and rhythmic movements, neonatal massage promotes physiological stability, improves weight gain, and supports neurobehavioral organization with no relevant adverse effects in preterm infants [4,5,6]. The neurobiological basis of these benefits lies in the activation of C-tactile (CT) afferents, a subclass of unmyelinated mechanoreceptors that respond optimally to slow, low-force dynamic touch. Experimental studies have shown that stroking at approximately 3 cm/s with light contact force of around 0.3 N increases parasympathetic activity and heart-rate variability in preterm infants, whereas static touch does not elicit comparable autonomic effects [7]. More recent work on affective touch, mainly through activation of tactile C-fibers, suggests that such stimulation promotes caregiver–infant bonding, reduces stress responses, and supports neurodevelopment [8].

Although massage therapy in the NICU has shown great potential, its use is limited due to personnel availability and inter-operator variability. Therefore, there is an unmet need for standardized, reproducible, and safe methods to deliver dynamic tactile stimulation. In this sense, soft robotic systems, e.g., soft pneumatic actuators, provide a promising pathway toward automated solutions that can mimic caregiver massage patterns while ensuring clinical safety and repeatability [9,10].

Supporting this direction, soft-robotics studies have demonstrated that pneumatic actuators generate controllable, human-safe contact forces with minimal electronics. In the context of affective touch, Zheng et al. recently introduced a soft-robotic haptic rendering system (S-CAT) that simulates CT-optimal stroking on the forearm by controlling velocity, normal force, and skin temperature, generating psychological and neurophysiological responses comparable to manual affective touch [11]. For example, fluidically programmed haptic textiles made from heat-sealable nylon–thermoplastic polyurethane (TPU) can reproduce programmable point-pressure cues at low internal pressures via simple thermal bonding [12]. In parallel, air-microfluidic actuators developed for lymphedema therapy and prosthetic socket liners achieve sequential inflation and uniform deflation through tailored channel resistances, producing graded compression profiles while reducing valve count, system size, and power consumption; these characteristics are essential for wearable use in fragile patients [13,14,15]. Additionally, passive flow-control networks and soft pneumatic circuits, including ring and relaxation oscillators, encode actuation timing directly within the fluidic architecture, eliminating the need for complex valve arrays and enabling lightweight, low-power control [16,17,18,19].

Although these technologies demonstrate that soft-pneumatic systems can deliver safe, repeatable contact forces, no existing neonatal device reproduces massage-like, sequential normal-pressure patterns for CT-afferent stimulation. For instance, current commercial or research systems focus on vibrotactile or static touch: for example Calmer simulates maternal heartbeat and respiratory motion to reduce procedural pain, and the Prapela SVS bassinet pad provides stochastic vibration for autonomic stabilization [20,21], and the recently developed Breathing Operator for BaBY (BOBBY) delivers brief, automated vibratory tactile stimulation in response to cardiorespiratory events in extremely preterm infants [22]. Neither, however, offers localized, low-magnitude, shear-free compression with programmable timing and amplitude. The present study, therefore, introduces a compact soft pneumatic actuator array (SPAA) capable of delivering sequential, and pressure-limited stimuli consistent with CT touch parameters, bridging clinical neurophysiology with engineering design to standardize neonatal massage therapy.

2. Materials and Methods

2.1. Design Specifications of the SPAA

The design of the SPAA was guided by neurophysiological, clinical, and safety considerations derived from studies on dynamic touch in preterm infants and soft-robotic haptic systems. The objective was to deliver shear-free, low-magnitude, sequential normal pressure within the range that activates C-tactile (CT) afferents. Figure 1 illustrates the design concept, placement, and actuation sequence of the SPAA.

2.1.1. Actuation Modality

By design, the SPAA applies shear-free normal compression to minimize epidermal stress in preterm skin. However, to balance this safety constraint with physiological relevance, it delivers slow, low-force pressure transients with second-scale ramps and inter-actuators inflation delays. Spatiotemporal sequencing activation across adjacent actuators reproduces a stroking-like percept via tactile apparent motion, consistent with pneumatic devices where shear-free normal indentations replicate the directional cues of lateral skin stretch [23,24,25,26]. Consistent with this idea, dynamic, low-force touch with a stroke velocity between 1 and 10 cm/s tuned to CT-optimized profiles is associated with increased parasympathetic activity and higher heart-rate variability in preterm infants, whereas static touch does not [7,27]. Taken together, this approach mimics stroke-like motion through sequential activation while adhering to dynamic, low-force touch profiles aligned with CT-afferent stimulation.

Furthermore, according to established studies on massage therapy for preterm infants, the standard protocol involves a 15 min session administered three times daily for 5 to 29 days with a mean duration of 15.61 days [28]. This session is divided into three 5 min phases: initial tactile stimulation, followed by kinesthetic stimulation, and concluding with tactile stimulation. The proposed SPAA is designed specifically to emulate the manual techniques required during the first and third phases [29].

2.1.2. Geometry and Contact Area

Each module of the SPAA measures 30 × 20 mm, and the array is tiled in a 1 × 4 matrix with 10 mm inter-chamber spacing, with the 30 mm edge oriented cranio-caudally. This yields a center-to-center pitch of 30 mm laterally and an overall active footprint of 30 × 110 mm. The 110 mm width aligns with typical internipple distances, ensuring coverage of the chest and upper abdomen with margin for a secure fit [30,31,32].

Given the lateral pitch of 30 mm between actuators, the sequential activation of modules can be used to emulate a stroking motion across the surface. To reproduce the clinically optimal stroke velocity for massage therapy (1–10 cm/s), the inter-actuator delay is set according to v = d/Δt, yielding delay intervals between 0.3 s and 3 s. This configuration ensures that the tactile stimulus is delivered with spatiotemporal characteristics tailored to effective neonatal massage therapy.

The SPAA dimensions (30 × 110 mm) were validated against anthropometric data for Late Preterm (defined as 34–36 weeks of gestation [33]) and Term infants. Unlike the flattened chest of adults, the neonatal thorax exhibits a cylindrical morphology often called “barrel chest”. Given this geometry, we analyzed the chest circumference (CC) to determine the available contact area. As summarized in

Table 1. Thoracic circumference and derived anterior arc availability.

Cohort	Gestational Age (Weeks)	Chest Circumference in cm (50th Percentile) [33]	Calculated Anterior Arc (cm)
Late Preterm	35–36	30	15
Term	37–42	33	16.5

, healthy late-preterm infants present a median CC of 30.0 cm [34]. This corresponds to an available anterior arc of approximately 15.0 cm (the width of the SPAA occupies approximately 73% of this arc). This ratio ensures broad pectoral coverage for stability while maintaining necessary clearance from the mid-axillary lines to facilitate respiratory movement.

Longitudinal fitting was evaluated using linear anthropometric reference values derived from a cross-sectional study of 732 newborns [35]. The mean sternum length ranges from 6.1 cm in preterm infants to 8.7 cm in term neonates. With a vertical dimension of only 30 mm, the SPAA fits entirely within the sternal region.

2.1.3. Force Target and Control Variables

Guided by Manzotti et al. [7], who showed that gentle, dynamic touch in preterm infants elicits parasympathetic activation at skin-level forces of ≈0.3–0.5 N, a target characterization window of 0.1–0.5 N was established per SPAA module to capture the effect range with a safety margin. By contrast, internal pneumatic pressure is treated merely as a control input to realize the calibrated force output and is not a therapeutic endpoint.

2.1.4. Preload and Attachment

Preload was applied using a Velcro belt set to 10, 20, and 30 g to emulate a soft garment fit. Prior to applying preload, the load cell was calibrated with certified weights, and the calibration was verified immediately before testing.

2.1.5. Thermal and Material Considerations

The SPAA is intended for use over the infant’s clothing while situated within an incubator to replicate the same scenario where massage therapy is given. Additionally, given that the environmental temperature is strictly regulated by the incubator, the low effective thermal conductivity of the nylon–TPU laminate (0.03–0.1 W·m⁻¹·K⁻¹) and the short, low-pressure duty cycles, conductive heat transfer at the actuator–skin interface is expected to be negligible; therefore, thermal effects were not modeled in this study [36,37].

In parallel, the nylon–TPU laminate was selected because of its flexibility and skin compatibility. Although the SPAA is intended for use over the infant’s clothing, the nylon–TPU fabric is classified as a surface device with prolonged contact with intact skin under the ISO 10993-1 classification for biological evaluation of medical devices [38]. Accordingly, in the worst-case scenario, the primary skin-contacting layer would be the Nylon 70 denier fabric with a single-sided TPU coating. Nevertheless, this material provides compliance for conformal contact and gentle pressure delivery while relying on a material with an established track record in the Neonatal Intensive Care Unit. TPU-coated nylon fabrics used in neonatal applications are certified to meet ISO 10993-5 (in vitro cytotoxicity) and ISO 10993-10 (skin irritation and sensitization) requirements [39], and the same nylon–TPU laminate is employed in FDA-cleared and CE-marked neonatal non-invasive blood pressure cuffs routinely used on preterm infants [40]. Thus, the chosen laminate combines the mechanical properties required for soft actuation with a biocompatibility profile consistent with existing NICU standard-of-care devices.

2.1.6. Sensing

Internal pressure was continuously monitored using a Honeywell 100PGAA5 transducer (Honeywell International Inc., Charlotte, NC, USA) and force was measured using a Mavin NA6 7 Kg load cell (Hope Technologic Co., Ltd., Xiamen, China).

2.2. SPAA Design

The SPAA was fabricated using two pieces of nylon–TPU-coated fabric with the TPU-coated sides facing inwards. This combination has been widely adopted in soft robotics because of its airtightness, flexibility, and mechanical robustness [9,10,12]. Each actuator consisted of a rectangular chamber measuring 30 × 20 mm (Figure 2a,b), dimensions selected to simplify heat-sealing and improve reproducibility while remaining proportional to the anatomical dimensions of preterm infants, thereby enabling localized stimulation over clinically relevant body areas.

In addition, each chamber was fitted with a 3D-printed polylactic acid (PLA) connector (Figure 2c) that serves as a coupling interface for standard IV tubing. This configuration allows direct integration with medical-grade tubing, ensuring reliable pneumatic sealing and compatibility with clinically available components.

2.3. SPAA Manufacturing

The actuators were manufactured using a thermal bonding process adapted from prior textile-based actuator studies [12,13,14,15], and the fabrication workflow is summarized in Figure 3. First, sheets of TPU-coated nylon fabric were segmented into an array of 4 modules of 30 × 20 mm, and a PLA connector was positioned in the center of each module to serve as the pneumatic inlet. Then, two fabric layers were bonded using a bag sealer at 110–120 °C, forming airtight chambers while preserving flexibility. This process is widely adopted in textile soft actuators due to its scalability, reproducibility, and simplicity [12]. To verify sealing, each actuator array underwent a submersion leak test where they were pressurized to 20–30 kPa for 60 s. If bubbles were seen, epoxy resin was applied to the holes.

This combination of heat-press sealing and 3D-printed connectors provided a balance between structural robustness, modularity, and manufacturability. Comparable fabrication strategies have been reported in wearable pneumatic lymphedema sleeves and prosthetic socket interfaces, where multi-chamber arrays generate controlled compression and massage sequences [13,14,15].

2.4. Actuator Sequencing

Two setups were tested to achieve a sequential actuation. First, in the parallel setup (Figure 4a), temporal sequencing was achieved by grading branch flow resistance using needle-valve speed controllers (SMC AS2002F-06A, SMC Corporation, Tokyo, Japan) placed at each manifold (Figure 4b) branch outlet. Specifically, each valve aperture was preset to a target resistance, which set the branch time constant and, therefore, produced predictable inter-actuator inflation delays. As a result, a progressive, SPAA wave-like activation was generated without active valves [15,16,17]. Meanwhile, outlet tubing lengths and diameters were matched to equalize hydraulic resistance and thus synchronize deflation and improve cycle-to-cycle repeatability.

On the other hand, in the series setup (Figure 4c), actuators were connected in cascade, and a speed controller (SMC AS2002F-06A) was installed at the outlet of each actuator, where each aperture was set to a defined position. Accordingly, by grading these setpoints along the chain, we imposed target inter-actuator delays and generated a traveling inflation wave without a manifold or active valves. Similarly, matched exhaust geometry thereby supported synchronized deflation and consistent repetition across cycles.

Taken together, this hybrid architecture combines passive fluidic sequencing through mechanical speed controllers to substantially reduce complexity, noise, and power versus valve-driven arrays, which is advantageous for neonatal settings. In addition, similar hybrid strategies have been reported in wearable soft-robotic sleeves and prosthetic socket liners, where simplified microfluidic designs replace heavy, power-intensive electronic control systems [12,13,14,15].

2.5. Manifold

In the parallel configuration, airflow was delivered via a Stereolithography (SLA) 3D printed manifold (Figure 4b) that evenly supplies all actuators. Clear V4 resin was chosen for its high-resolution internal features and smooth as-printed surfaces, thereby facilitating airtight channels and reliable sealing after post-curing. Moreover, to keep the design reproducible, the internal layout used simple geometric primitives compatible with standard SLA printers. Moreover, all outlet branches were matched in length and inner diameter to equalize pneumatic path resistance from the pump to each actuator, thus promoting uniform flow distribution.

2.6. Pressure Control System

The fluid actuation system operates using an open-loop control architecture. Unlike closed-loop systems utilizing active sensor feedback (e.g., PID), the current setup does not modulate pump output in real-time based on sensor data. Instead, the target stagnation pressure is established through a fixed-input regulation strategy combining electrical and mechanical controls.

Pressure regulation is achieved through a two-step process:

• Electrical Regulation: A programmable DC power supply is utilized to set limits on the pump’s input voltage (V). The DC power supply voltage was set to 2.0 V and is a step function with two states, 0 V and 2 V. This acts as a coarse control mechanism to establish the baseline hydraulic power and ensure the system operates within safe pressure limits.
• Mechanical Flow Control: Fine-tuning of the stagnation pressure is performed using manual in-line speed controllers (SMC AS2002F-06A). These variable throttle valves are adjusted manually to restrict the flow rate downstream of the pump. By altering the valve opening, hydraulic resistance is modified to achieve the precise target pressure required for each test case.

It is important, however, to mention the limitations inherent to this open-loop, manual approach. Because the system lacks an active feedback loop, it cannot automatically reject disturbances or compensate for dynamic pressure fluctuations during operation. The system relies on steady-state assumptions; therefore, precise manual calibration of the flow valves is performed immediately prior to data acquisition to ensure the target pressure is maintained within the desired tolerance.

For more details of the 3D printed parts, check the Supplementary Materials for more information.

3. Experimental Tests and Results

3.1. Experimental Setup

The textile pneumatic actuators were evaluated in two complementary setups. First, the setup in Figure 4a,c was used to measure internal pressure and time under both series and parallel routing using the same SPAA and instrumentation. In both setups, a Honeywell 100PGAA5 pressure sensor was integrated to record the internal pressure of each actuator during inflation, and the data was acquired using a 20 Hz sampling frequency and logged via an Arduino Uno microcontroller. In total, 3 measurements were taken per actuator to calculate the mean and SD of the curves. For a more detailed explanation of the setup, refer to Appendix A.

Second, Setup 2 (Figure 5a) was used to measure the normal force applied by a single actuator with a load cell and a preload using a Velcro belt. The load cell was a Mavin NA6 7 kg model, and it was calibrated with a 100 g calibration weight. Furthermore, a 3D-printed support was used to hold it, as seen in Figure 5b. The preload of 10, 20, and 30 g produced by the Velcro belt was measured in the same load cell before inflation. In addition, the same Honeywell pressure sensor was used to obtain internal pressure data, and 3 measurements were taken as well per preload to calculate the mean and SD of the curves.

3.2. Actuator Characterization

3.2.1. Characterization Curves

The actuator characterization was performed by mounting it on a load cell under controlled preloads of 10, 20, and 30 g, where cyclic inflations were executed to obtain data on pressure, force and time, from which characterization curves and time-domain kinetics were obtained.

Results of the characterization are shown in Figure 6. It summarizes the single-actuator characterization at preloads of 10, 20, and 30 g, reported as the mean ± SD over five cycles per preload, across three graphs: force–time, pressure–time, and force–pressure. In all cases, the pressure rose smoothly to ~100 kPa without overshoot, while the force increased to a plateau. Although the pressure is above the intended therapeutic range, they were included to obtain a limit characterization of the actuator behavior and to identify the onset of nonlinear deformation and proximity to material and seam failure. These measurements served as a stress test of the textile–TPU laminate and heat-sealed seams and were not intended to represent clinically relevant operating conditions.

Regarding the dynamic response, the rise time (measured as the interval between 10% and 90% of the saturation pressure) improved with higher preloads, decreasing from 1.8 s at 10 g to 1.2 s at 30 g. Similarly, the settling time (defined as the time required to stay within 2% of the 100 kPa plateau) ranged between 3.2 and 4.1 s across preload conditions. Moreover, the 10 g condition reached 30 kPa in <0.5 s and achieved a peak force of approximately 0.47 N at approximately 3.5 s; the 20 g condition reached 30 kPa in <0.10 s and achieved a peak force of approximately 1.53 N at approximately 2.0 s; and the 30 g condition reached 30 kPa in <0.06 s and achieved a peak force of approximately 1.66 N at approximately 3.5 s. Collectively, these results indicate that higher preload shifted the entire force trace upward with minimal change in shape, inflation dynamics are stable and non-oscillatory, and a nonlinear relationship between force and pressure.

The force–pressure relation established here shows that, for our actuators, an internal-pressure window of approximately 20–60 kPa produces contact forces of about 0.1–0.3 N. Accordingly, in the remainder of the paper, we specify and evaluate performance in terms of pressure within this verified operating band.

3.2.2. Actuator Pressure According to Speed Controller Position

After achieving a relation between force and pressure, we focused on understanding how different configurations of the speed controller affect the inflation of the actuator. Thus, speed control was performed on a single actuator of the SPAA by systematically varying the outlet resistance (SMC AS2002F-06A knob position) and recording the resulting chamber pressure–time response for each setting. Knob position indexing follows the convention R0 = fully closed, and from R1 onward each step corresponds to an additional 45° of opening (R1 = +45°, R2 = +90°, …), with higher indices implying lower resistance. The maximum opening that the speed controller has is R96.

Figure 7 shows pressure vs. time responses of four different speed controller positions (R4, R5, R6, R8), where solid traces are cycle means and shaded bands denote ±1 SD. In all cases the signal rises from approximately 0 kPa toward a quasi-steady plateau; at R8 the curve reached 20 kPa by approximately 0.7 s and settled near 55–57 kPa; at R6 it reached 20 kPa by approximately 2.4 s and plateaus around 43–46 kPa; at R5 it reached 20 kPa by approximately 3.8 s with a plateau of 33–36 kPa; and at R4 the response was the slowest, approaching 18–20 kPa at approximately 20 s. The dominant differences across settings were the final pressure level and the time required to reach the maximum pressure. Collectively, these results indicated repeatability and that higher controller positions produced progressively faster inflations and higher steady levels.

For this study, the specific controller positions shown in Figure 7 (R4, R5, R6, R8) were chosen to keep the actuator within the approximately linear region of its force–pressure characteristic and within the target therapeutic pressure band for preterm infants. In practice, R8 was empirically identified as the highest setting that produced a plateau pressure of ≈55 kPa, beyond which the force–pressure relationship clearly became nonlinear and the resulting contact forces exceeded the desired range. Therefore, progressively more restrictive settings were selected (R6, R5, R4) by increasing the resistance to obtain a family of pressure–time curves with lower steady pressures and longer inflation times, while still spanning the 20–55 kPa interval relevant for safe, massage-like stimulation of the preterm thorax. Higher opening indices were not investigated further, as they yielded pressures outside this regime and were not compatible with the intended neonatal application.

3.3. SPAA Pneumatic Network

3.3.1. Parallel Setup

To test the control system, both of the setups explained earlier were tested. First, the parallel setup was tested, and as mentioned previously, only the parallel configuration used a common manifold that fed the four actuators of the SPAA, each of which used an SMC AS2002F-06A speed controller to impose different resistances. Furthermore, synchronized pressure–time recordings were taken to quantify inner pressure and to compare actuator-averaged responses.

Figure 8a compares actuators A1 to A4, each with four speed controller settings (A1_R22, A2_R14, A3_R10, A4_R8); solid traces are cycle means and shaded bands indicate ±1 SD. A1_R22 reached 20 kPa at approximately 0.27 s and settled at the highest plateau; A2_R14 reached 20 kPa by approximately 0.38 s and plateaued below R22; A3_R10 rose to approximately 20 kPa at approximately 0.78 s and approached a 31–32 kPa plateau at approximately 1.7 s; and A4_R8 was the slowest and reached the lowest pressure, as it reached 20 kPa at approximately 2.27 s and plateaued near 21–22 kPa. Variability on the plateau was small, indicating good repeatability. Furthermore, the primary differences across settings were the final pressures attained and the time required to reach them, and increasing the setting from R8 to R22 yielded progressively faster inflations and a higher plateau.

The sequential activation was verified by comparing the time each actuator took to reach 20 kPa (Figure 8b). The order was the following: A1 (0.27 s), then A2 (0.38 s), then A3 (0.78 s), and finally A4 (2.27 s). The dead time, time in seconds that it took to reach 20 kPa between actuators, was about 0.11 s (A1–A2), 0.4 s (A2–A3), and 1.49 s (A3–A4), they were larger than the variability indicated by the shaded bands, confirming not only a clear, programmed sequence from A1 to A4 but also a concise behavior between replications. This staggered timing is achieved solely by adjusting the per-branch speed controller.

3.3.2. Series Setup

In the series configuration, the actuators were arranged in a chain series: the air pump fed Actuator 1 (A1) directly; subsequently, A1’s outlet passed through a speed controller into Actuator 2 (A2), and so on. By progressively adjusting the speed controller knob along the chain, inter-actuator delays (Δt) were imposed and thus generated a sequential inflation pattern. Meanwhile, time–pressure data from each chamber were used to quantify the inflation time delays between actuators and to verify cycle-to-cycle repeatability.

Figure 9a shows the mean pressure–time traces for actuators A1–A4 in the series configuration. All curves started at 0 kPa and rose toward plateaus. A1 (first in the chain) was the fastest and highest, reaching 20 kPa at 0.55 s and settling within approximately 1 s at 50–52 kPa, where it remained for the observation window. A2 rose at a slower rate and plateaued at a lower value, crossing 20 kPa at approximately 0.96 s. A3 was even slower as it approached 20 kPa at approximately 2.22 s. A4 (last in chain) was the slowest and lowest, climbing to approximately 20 kPa at 3.44 s.

Sequential activation is evidenced from the non-overlapping times each actuator took to reach 20 kPa (Figure 9b): A1 at 0.55 s, A2 at 0.96 s, A3 at 2.22 s, and A4 at 3.44 s. The dead time, time in seconds, that it took to reach 20 kPa between actuators, is, therefore, about 0.41 s between A1 and A2, 1.25 s between A2 and A3, and 1.22 s between A3 and A4, for a total spread of about 2.88 s from A1 to A4. These progressively larger gaps downstream are characteristic of series routing, and they visually confirm a clear A1 → A2 → A3 → A4 activation sequence.

To guarantee replicability in the series configuration, three independent SPAA prototypes were fabricated and tested. Each prototype underwent five consecutive inflation–vent cycles, resulting in a total dataset of fifteen cycles per actuator. In Figure 10, each subfigure has thin traces that represent the three recordings of each actuator; the thick blue line represents the mean across them, and the light band shows ±1 SD. Plateaus decreased along the chain (A1 > A2 > A3 > A4): A1 settled near 50–52 kPa with a small oscillatory ripple, A2 near 41–42 kPa, A3 around 29–31 kPa, and A4 around 25–26 kPa. Across actuators, the inflation/deflation windows aligned closely, the waveforms were stable over repetitions and the narrow SD bands indicated low cycle-to-cycle variability, confirming that the protocol yielded consistent dynamics within and across recordings.

In Figure 11, the plot overlays the cycle-averaged pressure profiles for A1–A4 across three inflation intervals, where solid curves are the means of the data of each actuator and the shaded bands indicate ±1 SD. Waveform shapes were consistent across cycles, and plateaus followed the series order: A1 was the highest, then A2, then A3, and A4 was the lowest. A1 rose rapidly and settled near 50–52 kPa with a small ripple; A2 plateaued around 41–42 kPa; A3 around 29–31 kPa; and A4 around 25–26 kPa. Variability was small on the plateaus and briefly larger during venting, most noticeably for A4. Overall, the figure showed stable, repeatable cycles and a clear upstream-to-downstream gradient in both final pressure and time to approach it.

Table 2. Time to reach 20 kPa for each actuator in the parallel and series configurations.

Actuator	Parallel Configuration		Series Configuration
	Time to 20 kPa (s)	STD	Time to 20 kPa (s)	STD
A1	0.27	0.049	0.555	0.036
A2	0.38	0.026	0.962	0.037
A3	0.78	0.084	2.216	0.168
A4	2.27	0.156	3.438	0.084

summarizes the time each actuator took to reach 20 kPa in the parallel and series configurations, complementing the results shown in Figure 8 and Figure 9.

3.4. Demonstration of Sequential Activation

Sequential actuation was demonstrated on a StandInBaby^® mannequin approximating the size and mass distribution of a 50th-percentile term newborn. Figure 12a shows the SPAA placement. Figure 12b presents the time-staggered chamber activation; the timestamp indicates the elapsed time since inflation initiation for the depicted frame. It should be noted that these mannequin trials were conceived solely as a qualitative, visual proof-of-concept to show that the SPAA can conform around an infant-sized thorax and achieve sequential activation on a curved, anatomically relevant surface. They are not intended as quantitative validation, which will require instrumented, preterm-appropriate phantoms to characterize interface pressures and force distributions in future work.

4. Discussion

Massage therapy is an evidence-based intervention to address the lack of tactile stimulation that preterm infants experience during NICU stays. It has been shown to promote physiological stability, improve weight gain, and support neurobehavioral organization [4,5]. In this context, soft robotics has emerged as a suitable option. Studies [12,13,14,15] demonstrate that it is possible to deliver stimuli with reproducible, safe normal forces (0.3–0.5 N) [7] and programmable sequential pressure patterns. However, a clinically integrated neonatal device of this kind is still lacking since current options range from vibrotactile stimulation to static touch.

Given these needs, the present study introduces an experimental validation of a soft-pneumatic actuator array (SPAA) capable of delivering controlled stimuli. The SPAA was fabricated from nylon fabric laminated on one side with TPU. This material was selected due to its airtightness, flexibility, and mechanical robustness [9,10,12]. The dimensions were set to 30 × 20 mm to simplify heat-sealing, improve reproducibility, and keep the SPAA proportional to preterm-infant anatomy. Subsequently, tests were performed on a single actuator to find a relationship between internal pressure and normal force. The results indicated that the pressure–time responses were smooth and the force–pressure curve (Figure 6) was almost constant between 0 and 50 kPa and increased rapidly at higher pressures. Then, attention was directed to the achievement of sequential actuation. A series of speed-controller configurations were systematically evaluated to test their influence on actuator inflation (Figure 7). The findings demonstrated that inflation could be controlled exclusively through the adjustment of the speed controller setpoint. Therefore, two multi-actuator configurations (Figure 4) were tested with the objective of reproducing the spatiotemporal sequence characteristic of neonatal massage therapy.

In the series configuration, analysis of inflation time across different runs (Figure 9, Figure 10, Figure 11 and Figure 12) proved clear, programmable inter-actuator delays with low cycle-to-cycle variability, thereby enabling a traveling-wave pattern from a single pump without active valves. Importantly, these delays correspond to virtual stroke speeds within the 1–10 cm/s range required for optimal CT-afferent stimulation in neonatal massage therapy, ensuring the actuator sequence delivers physiologically relevant tactile stimulation. By contrast, the parallel configuration achieved similar plateau pressures, but sequencing was inconsistent across trials. Although actuator dimensions were standardized, tube lengths were equalized, and the system was repeatedly calibrated, measurable variability persisted. This was observed consistently when operating with low-flow regimes, where small knob tolerances and mechanical hysteresis produced disproportionately large changes in effective resistance. This variability was not observed in the series configuration. A plausible explanation is that the single flow path through the series chain attenuated the needle valve’s low-flow variability, thus small resistance differences had less effect on timing. Accordingly, the series configuration was identified as the optimal configuration for controlling sequential activation in this study. However, if the replicability limitation of the parallel configuration were addressed, for example, by using a different model of speed controller, then it would likely be preferable to the series configuration, as each actuator’s restriction could be controlled independently.

From a fluidic-resistance standpoint, the series configuration behaves as a single, distributed Resistance-Capacitance line in which resistances add along a single flow path, so local fluctuations in valve setting or actuator geometry are averaged over the chain and produce small, proportional shifts in all activation delays without altering the activation order, rather than abrupt changes in sequence. In contrast, the parallel network requires that multiple branches share a common source of pressure. Taking that into account, then small differences in branch resistance (e.g., due to needle-valve hysteresis or tubing tolerances) are exponentially amplified, diverting flow toward the lower-resistance paths and destabilizing the intended timing. This also affects flow distribution: in series, each actuator temporarily receives the full available flow until it reaches its maximum pressure, whereas in parallel, early-inflating actuators can transiently reduce driving pressure for the others, further degrading synchronization. Consequently, the series architecture exhibits higher intrinsic system stability to perturbations in supply pressure, valve position, and downstream loading, while the parallel configuration is more sensitive but, if equipped with highly precise and low-hysteresis flow controllers, could offer finer, independent tuning of each actuator’s pressure–time profile.

Based on the force–pressure characterization, it has been demonstrated that the applied force can be controlled through pressure setpoints and, in the same manner, pressure setpoints can be controlled through different speed controller configurations. Our operating range, set between 20 and 50 kPa, produced approximately 0.1–0.3 N of force, which lay within the target characterization window of 0.1–0.5 N stated previously. This range corresponded to the required for CT-afferent stimuli [7] while maintaining a safe margin. To confirm sequential activation, measurements of the time each actuator took to cross a 20 kPa threshold were taken and reported in Table 2. Moreover, additional measurements up to 100 kPa were performed exclusively to define mechanical limits and safety margins for the actuators and should not be interpreted as candidate operating pressures in a clinical setting.

Overall, the pressure–force calibration and timing measurements showed that valve settings allowed control of rise time and duty cycle; therefore, stimuli can be delivered at clinically relevant speeds while remaining under pressure limits. In other words, the system’s controllable pressure domain coincided with the forces needed for dynamic touch [7], and the series routing supplied the spatiotemporal structure to deliver it.

Compared with prior textile and pneumatic wearables, the SPAA operates at internal pressures (≈20–50 kPa) and normal forces (≈0.1–0.3 N), comparable to, or below, ranges reported for fluidically programmed haptic textiles, air-inflated sleeves, and pneumatic tactile displays (12.5–100 kPa; 0.35–9.8 N) [12,13,14,15,26]. Unlike systems that depend on multi-valve manifolds, embedded sensing, or microcontroller-based timing, the SPAA sequencing is passively encoded via tuned flow resistances (parallel) or graded outlet restrictions (series), consistent with electronics-free pneumatic oscillators and mechanically programmed logic [16,17,19,41]. This architecture reduces hardware complexity while inherently limiting peak loads, supporting safety for fragile neonatal skin.

Several limitations were observed during the development of this work. First, the results obtained were based on laboratory bench tests, and a single-actuator force–pressure characterization was used to infer force applied by measuring pressure; thus, validation on curved and compliant neonatal skin remains to be performed. Future work will incorporate thin-film or textile-based sensors integrated into compliant phantoms that better mimic neonatal thoracic tissue, enabling more realistic mapping of both normal and shear interface pressures. Furthermore, in this study, a 1 × 4 layout was evaluated; thus, a complete 4 × 4 actuator grid would be required in order to confirm spatial coverage of a preterm infant’s belly. From a safety standpoint, protections were implemented in software; however, clinical use will require hardware pressure-limit devices and explicit bounds on cumulative load per cycle and session (e.g., an inline relief valve or a pressure switch that automatically cuts pump power above a set threshold). In addition, in terms of the pressure control system, it lacks an active feedback loop; thus, it cannot automatically compensate for dynamic pressure fluctuations during operation. Also, the system relies on steady-state assumptions; therefore, precise manual calibration of the speed controllers must be performed to ensure that the target pressure is maintained within the desired tolerance. Furthermore, future work should include a closed-loop system that regulates the actuator’s internal pressure to the infant’s physiological indicators of stress. Regarding control and automation, the current prototype can be operated in an open-loop control to automate activation; however, in the absence of embedded on-body sensing, the delivered pressure/force may vary with garment fit. As a mitigation, enforcing conservative duty-cycle limits in software is recommended, but future clinical versions should incorporate closed-loop regulation (e.g., a low-profile pressure transducer or thin-film force/pressure sensor) to keep each actuator’s applied force within the design specifications. In addition, it should be acknowledged that the present work focuses on experimental validation only. Future work will include analytical and simulation models to optimize control.

In terms of shear forces and interface mechanics, although the present system was designed primarily to generate controlled normal pressures on the thoracic surface, we acknowledge that shear forces may still arise under real-use conditions. Two potential sources are particularly relevant: (i) tangential components of the strap tension along the skin surface, especially when the Velcro strap is tightened asymmetrically, and (ii) local misalignment between the actuator’s expansion direction and the local surface normal on the curved and highly compliant neonatal thorax. In future iterations of the device, shear loads may be mitigated by using wider and more compliant straps to distribute tension, by introducing low-friction or low-adhesion textile layers at the skin–device interface to reduce lateral drag, and by integrating the actuators into a vest-like garment that minimizes relative sliding between the actuators and the skin. Quantitative measurement of shear forces was beyond the scope of this initial engineering study but represents an important objective for future work, ideally in combination with anatomically realistic phantoms or in vitro models of neonatal skin.

Although the SPAA dimensions (30 × 110 mm) were validated using late-preterm measurements, it is important to consider the natural differences in body size among NICU infants for future designs. Data from Al-Samarrai et al. show a key limitation regarding torso length: while term infants have an average length of 19.8–20.0 cm, preterm infants average only 17.0 cm [34]. This 3.0 cm difference is substantial; it implies that expanding to a larger matrix (e.g., 4 × 4 grid with 12 cm vertically) would cover more than 70% of a preterm infant’s torso, potentially interfering with critical areas such as the neck or umbilical cord

Therefore, while the current SPAA is a compact proof-of-concept that fits safely within the sternal zone (mean length 6.1 cm [34]), future versions must stay within the 17.0 cm limit, ensuring safe clearance for the umbilical cord and neck movement. Regarding width, the variability is also high. Siyah Bilgin et al. report that chest circumference in late-preterm infants ranges from 29.0 to 32.0 cm, compared to the 33.0 cm median of term neonates [33]. This variation is effectively managed by the flexible design of the nylon–TPU actuators; the SPAA’s soft structure allows the 110 mm width to adapt to different chest curves, ensuring effective contact during chest movements caused by breathing or crying and physical differences typical of preterm infants, such as abdominal swelling or edema.

5. Conclusions

This work validates a compact SPAA system that delivers shear-free, massage-like compression within the neurophysiologically relevant range for preterm infants (0.1–0.3 N; 20–50 kPa). Among the two architectures examined, the series configuration produced repeatable inter-actuator delays and a traveling pressure wave (A1–A4). In contrast, the parallel configuration showed poor reproducibility at low-flow conditions across tests. Collectively, these results demonstrate that passive soft-pneumatic sequencing can be leveraged to standardize and automate tactile stimulation, supporting consistent therapeutic protocols suitable for neonatal care treatments. The proposed system establishes a safe, simple-to-use, and reproducible platform for dynamic touch and offers a promising alternative to operator-dependent manual massage. Moreover, because a nylon–TPU fabric was used, the SPAA is easy to manufacture; thus, it can effortlessly adapt to infant dimensions and require only minor flow-resistance adjustments to remain within the design specifications.

Future studies should extend this work beyond bench testing to quantify pressure and shear distributions on anatomically realistic, compliant neonatal thoraces and to verify comfort and skin integrity during prolonged use. In parallel, the pneumatic network could be refined, e.g., by leveraging microfluidic flow resistors and alternative layouts to tailor sequencing delays, pressure gradients, and coverage to different gestational ages, body weights, and thoracic morphologies. Integrating thin-film pressure or force sensors into the SPAA would enable closed-loop, infant-specific control of stimulation patterns and facilitate real-time safety monitoring in the NICU. Finally, preclinical and clinical studies are needed to determine whether the standardized, shear-free massage delivered by this platform can modulate arousal, cardiorespiratory stability, stress biomarkers, and ultimately neurodevelopmental outcomes in preterm infants, compared with conventional handling and manual massage.