MELEGROS: Monolithic Elephant-Inspired Gripper

• MELEGROS is a monolithic, soft gripper built as a single 3D-printed lattice that integrates structure, pneumatic actuation, and optical sensing.
• It employs five independent pneumatic subsystems to enable continuum motions such as reaching, bending, pinching, and scooping, mimicking elephant trunk kinematics.
• The embedded optical waveguide sensors decouple tactile, bending, and actuator signals, providing phase-specific sensory feedback for precise manipulation.

MELEGROS, short for Monolithic ELEphant-inspired GRipper with Optical Sensors, is a fully soft, pneumatically actuated, elephant-trunk-inspired gripper in which the body, actuators, and sensors are all the same 3D-printed material and are fabricated in a single print (Trunin et al., 24 Sep 2025). Its central premise is that actuation and multi-modal sensing can be co-fabricated within one continuous soft lattice structure, rather than assembled from mechanically dissimilar materials or post-integrated sensor skins. The resulting device combines a continuum body, five independent pneumatic subsystems, and six embedded optical waveguide sensors to perform trunk-like reaching, bending, pinching, scooping, and delicate grasping, while functionally separating tactile and proprioceptive sensing.

1. Biological model and design rationale

MELEGROS is modeled on the distal morphology of the African elephant trunk, whose distal part has two asymmetric “fingers” that can pinch, scoop, and support objects. In that biological template, the tip and the arm are not separated mechanically or functionally, and control relies heavily on distributed tactile and proprioceptive feedback, allowing dexterous manipulation even when vision is occluded (Trunin et al., 24 Sep 2025).

The design follows that template in three stated ways: a continuum body with no rigid joints, intrinsic sensing in which the body material itself forms optical waveguides, and integration of the proximal “arm” and distal “fingers” as one continuous structure. This organization distinguishes MELEGROS from architectures in which a soft gripper is attached to a separate arm or in which sensing is added as an external layer.

The paper frames this approach against several limitations of prior soft grippers. Multi-material capacitive or resistive sensors commonly require conductive fillers or separate conductive layers that are mechanically stiffer than the bulk elastomer and introduce interfaces prone to delamination, fatigue, and modeling uncertainty. Post-assembly sensing complicates fabrication and calibration and makes coupled mechanics harder to simulate or predict. Tendon-based systems provide rich kinematics but require routing cables and often rigid terminations, while sensing is usually separate and external. Even recent “somatosensitive” actuators may rely on multiple materials and complex fabrication workflows and are therefore not truly monolithic in the sense used here.

Within this context, “monolithic” has a precise meaning. It denotes both single material and single continuous print: the lattice body, pneumatic membranes, internal cavities, and optical waveguides are all printed from one transparent soft resin in one SLA job. A common misconception is that monolithic soft robots are merely jointless or highly integrated. In MELEGROS, the term refers more specifically to elimination of internal material interfaces between structure, actuation, and sensing.

2. Morphology, lattice architecture, and fabrication

The body of MELEGROS is a triply periodic minimal surface lattice of IWP type rather than a solid elastomer (Trunin et al., 24 Sep 2025). The IWP-TPMS has 12.5 mm unit cells and 1.5 mm wall thickness. It is selected for low bending stiffness relative to a solid of the same envelope, high compliance and lateral deformability, and internal paths and nodes that support both pneumatic chambers and optical waveguides without sacrificial supports during printing.

Experimentally, 62.5 mm cubic samples, corresponding to $5 \times 5 \times 5$ cells, are compressed to 20–60% strain and show nonlinear but recoverable behavior. For FE simulations, the response is homogenized and linearized up to 40% strain with an effective Young’s modulus

$E_{\text{eff}} \approx 12\ \text{kPa}.$

This homogenized lattice serves as fabrication scaffold, contact interface, and transmission medium for actuator motion. The paper also notes a practical geometric consequence: the approximately 8 mm void inside each 12.5 mm lattice cell essentially sets the minimum graspable object size, because smaller objects may slip through rather than be retained.

The global geometry contains a proximal region functioning as an “arm” and a distal region with two asymmetric fingers. Pneumatic actuators are implemented as bladder-like chambers with circular cross-sections arranged in series and half-embedded into the lattice, so that lattice attachment constrains deformation and converts membrane expansion or contraction into bending or combined modes. The distal tip contains one dorsal finger actuator with 6 chambers and one ventral finger actuator with 4 chambers. The proximal region contains three 6-chamber actuators: one dorsal actuator along the central axis and two ventral actuators forming a ventral ridge.

Fabrication uses a Formlabs Form 4 SLA printer and Elastic 50A Resin, with 0.1 mm layer thickness, an energy dose of 38.40 mJ/cm² for perimeter, model, and support regions, and light intensity of 11.5 mW/cm². Post-processing consists of a 20 min IPA wash in Form Wash and a 70 °C, 30 min cure in Form Cure. Because the TPMS architecture provides in situ support paths and avoids trapping liquid resin, the lattice, actuators, and waveguides are printed without internal supports.

These fabrication choices have several stated consequences. The absence of material discontinuities removes stiff inclusions and adhesion interfaces, simplifies fabrication by avoiding casting and sensor insertion, and improves predictability of simulation because the constitutive behavior of sensors, actuators, and body is identical.

3. Pneumatic actuation and simulation-guided design

MELEGROS is powered by five independent pneumatic actuators: three proximal 6-chamber bladders and two distal finger bladders, with 6 chambers dorsally and 4 ventrally at the tip (Trunin et al., 24 Sep 2025). Positive and negative electronic pressure regulators, specifically SMC ITV-0010 and ITV-0090, are used together with 3/2 and 2-port solenoid valves per bladder. Pressure is commanded through analog outputs from an NI USB-6218 I/O device controlled in LabVIEW, over an operating range of approximately $-20\ \text{kPa}$ to $+50\ \text{kPa}$ relative to ambient.

Different pressure combinations produce distinct continuum motions. Pressurizing all three proximal actuators together yields reaching or axial extension. Applying negative pressure down to about $-20\ \text{kPa}$ produces axial contraction and often finger opening. Selective pressurization of dorsal or ventral proximal actuators produces dorsal or ventral bending of the arm. At the tip, the fingers can close symmetrically toward the central grasping axis or asymmetrically, for example with dorsal scooping while the ventral finger remains passive.

Actuator characterization is performed on half-embedded actuators used as a proxy for finger segments. Under free bending, pressure from $-20\ \text{kPa}$ to $+50\ \text{kPa}$ produces measured angles from about $-50^\circ$ to $75^\circ$. Under $+50\ \text{kPa}$, the same configuration produces a blocking force of 1.25 N. The pressure–angle relation is described as monotonic and approximately nonlinear.

The design workflow is simulation-driven. A full continuum model is built in SOFA, with geometries imported as STEP files, tetrahedral meshing through Gmsh, and merger into a single unstructured grid. The homogenized lattice constitutes the bulk envelope; pneumatic membranes are represented as separate deformable regions mapped to that envelope; internal cavities define surfaces where pressure is applied. Waveguides are modeled as stiffer subdomains to capture the stiffening effect of sensor inclusion.

Two simulation stages are distinguished. Model A is a sensor-position model that tracks candidate regions without explicit sensor bodies and evaluates point trajectories and local angles during actuation cycles. Model B is a sensor-integrated model that includes the waveguides explicitly and re-simulates to confirm that sensor placement does not degrade mechanical behavior and that bending occurs primarily where desired. The paper states two design criteria: tactile waveguides should experience less than $E_{\text{eff}} \approx 12\ \text{kPa}.$0 bending during normal finger bending, and adding waveguides should change global bending angles by less than $E_{\text{eff}} \approx 12\ \text{kPa}.$1. Simulated sensor angles of about $E_{\text{eff}} \approx 12\ \text{kPa}.$2 for bending sensors were considered sufficient for robust optical response.

The paper does not provide a closed-form kinematic model, but it conceptually writes segment curvature as

$E_{\text{eff}} \approx 12\ \text{kPa}.$3

with $E_{\text{eff}} \approx 12\ \text{kPa}.$4 obtained from FE simulation or empirical calibration and $E_{\text{eff}} \approx 12\ \text{kPa}.$5 the segment length. A plausible implication is that the monolithic constitutive continuity makes homogenized FE modeling unusually tractable for soft manipulator co-design.

4. Embedded optical sensing and signal decoupling

A defining feature of MELEGROS is the integration of six optical waveguide sensors directly into the soft lattice and actuator walls, using the same resin as the rest of the structure (Trunin et al., 24 Sep 2025). Each sensor is a 3D-printed polymer waveguide with a patterned surface consisting of wells intended to increase sensitivity to bending. The waveguide thickness is 1.5 mm. The wells have depth $E_{\text{eff}} \approx 12\ \text{kPa}.$6 and width $E_{\text{eff}} \approx 12\ \text{kPa}.$7, dimensions selected via COMSOL simulations to optimize linearity and sensitivity while remaining printable at approximately 50 µm printer resolution.

The optical transduction chain uses an IR LED, VSMY1850 with peak near 850 nm, and an IR photodiode, VEMT7100X01. Light propagation relies on total internal reflection; bending or local deformation at the patterned surface increases scattering or leakage and reduces transmitted intensity. The photodiode output voltage $E_{\text{eff}} \approx 12\ \text{kPa}.$8 is measured on a custom PCB, acquired in Python, and analyzed off-line. The paper conceptually describes sensor response as

$E_{\text{eff}} \approx 12\ \text{kPa}.$9

where $-20\ \text{kPa}$0 is local curvature, $-20\ \text{kPa}$1 strain, $-20\ \text{kPa}$2 contact force, and $-20\ \text{kPa}$3 nearby chamber pressure, while noting that no explicit analytical form is fitted.

The six sensors are organized as three per finger. Two tactile sensors are placed on the inner surfaces of the fingers and are approximately 37.5 mm long. Two actuator sensors, approximately 25 mm long, are placed on the last chamber of each finger actuator to measure local expansion or contraction of the terminal bladder. Two bending sensors run along the finger lengths, with lengths matched to the asymmetric fingers and approximately 37.5 mm on the shorter configuration. Thus each finger contains a tactile sensor, an actuator sensor, and a bending sensor, all embedded in the same printed material.

The placement strategy is explicitly aimed at decoupling exteroception from proprioception. Bending sensors are located in regions expected to experience high and consistent curvature but minimal external contact. Tactile sensors are located where object contact is likely but where pure finger bending should produce little signal. Actuator sensors are tightly coupled to chamber expansion and intended to respond primarily to local pressure-induced deformation.

Empirical characterization on a sensorized half-embedded actuator supports this decoupling. In a blocking-force test up to 50 kPa, the tactile sensor exhibits the largest voltage change, approximately

$-20\ \text{kPa}$4

while the bending sensor changes by about $-20\ \text{kPa}$5 and the actuator sensor by about $-20\ \text{kPa}$6. In free bending from $-20\ \text{kPa}$7 to $-20\ \text{kPa}$8, the bending sensor becomes dominant, with

$-20\ \text{kPa}$9

and the tactile and actuator sensors showing much smaller changes. All signals are reported as consistent over $+50\ \text{kPa}$0 repeats, and additional cyclic tests over $+50\ \text{kPa}$1 cycles suggest good repeatability with low drift.

These results establish functional rather than absolute decoupling. The tactile sensor is most responsive to external contact, the bending sensor to global shape change, and the actuator sensor to local chamber deformation. The paper does not report explicit SNR values or fitted calibration curves, but it interprets the observed voltage ranges and repeatable waveforms as evidence of distinct sensing modalities within one continuous soft body.

5. Manipulation repertoire and experimental demonstrations

MELEGROS weighs 132 g and is reported to lift a cylindrical object weighing 264 g, with radius 15 mm and length 120 mm, which is more than twice its own weight (Trunin et al., 24 Sep 2025). Demonstrated motions include reaching, pinching, scooping, dorsal and ventral bending, and multi-object enveloping. Because the proximal arm and distal fingers are one continuous structure, the system can approach objects from multiple directions without requiring precise pre-alignment.

Grasping trials include spheres, cubes, and stars with characteristic sizes of 12.5 mm and 25 mm, as well as grapes. In a representative UR5e-mounted pick-and-place protocol, the device begins from a relaxed, contracted configuration; proximal actuators contract and fingers open; the arm approaches the objects; the structure extends and fingers close to grasp; the arm moves above a receptacle; and the gripper opens to release. The lattice supports multi-object grasps, including the reported example of grasping three 12.5 mm star-shaped objects by enveloping them within the lattice and closing the fingers.

The sensory signatures recorded during these tasks are phase-specific. Actuator and bending sensors show consistent patterns during axial contraction and opening, dorsal and ventral base bending, and finger closing or elongation. Tactile sensors respond sharply when the lattice or finger presses against objects. For large objects such as 25 mm stars, both tactile sensors activate; for intermediate 12.5 mm cubes, only the ventral tactile sensor may activate depending on object position; for 12.5 mm spheres, no tactile sensors activate because the spheres are captured by lattice voids without compressing the finger surfaces. This distinction is central to the claim that the sensing suite can separate touch from motion and from internal pressure.

The grape-picking experiment is the most detailed demonstration of delicate manipulation. The sequence begins with the device contracted and closed; the fingers open; the base performs dorsal bending toward the cluster while distal sensors remain flat because only proximal actuators move; the arm reaches by elongating ventral actuators; the dorsal finger approaches a specific grape; fingertip contact produces a clear tactile peak; finger closure pinches the grape; detachment from the stem produces a higher tactile peak; and the signal drops after the grape is removed and contact is lost. The device then bends ventrally and opens to place the grape. The paper uses this sequence to show that even under open-loop pressure commands, the embedded sensors generate localized contact events and interpretable motion phases.

6. Position within soft robotics, limitations, and future directions

The paper identifies several contributions of MELEGROS within soft robotics (Trunin et al., 24 Sep 2025). First, it presents a monolithic elephant-inspired gripper with co-fabricated optical sensors: a single-material, fully 3D-printed lattice structure integrating five pneumatic actuators and six optical waveguide sensors. Second, it introduces a simulation-reinforced monolithic design workflow based on homogenized FE simulation in SOFA, with specific emphasis on sensor placement and decoupling of tactile and proprioceptive signals. Third, it uses an IWP-TPMS lattice both to enable support-free fabrication of internal cavities and to provide compliant trunk-like kinematics. Fourth, it demonstrates multifunctional perception through distinct optical responses for touch, bending, and chamber deformation. Fifth, it shows a payload of twice the system weight without rigid reinforcement or hybrid reinforcement strategies.

This positioning also clarifies what the system does not yet claim. No explicit feedback controller is implemented, no machine learning model is trained, and no quantitative mapping from voltages to force or curvature is deployed in closed loop. Sensor data are used primarily for characterization and demonstration, although the paper notes that the signatures are rich enough for phase detection, contact detection, and object-size or object-location inference. The conceptual relation

$+50\ \text{kPa}$2

is presented as a possible future direction rather than an implemented estimator.

The limitations are stated directly. Durability and fatigue are only briefly addressed through 100-cycle testing. The present device is at trunk-tip scale, and scaling toward a full trunk would require larger build volumes or modular assembly, potentially together with graded materials or interpenetrating lattices. The current simulation strategy relies on homogenization and linearized stiffness and does not include more detailed nonlinear or fluid-structure formulations. Control and calibration remain open problems, including systematic $+50\ \text{kPa}$3–$+50\ \text{kPa}$4 and $+50\ \text{kPa}$5–$+50\ \text{kPa}$6 calibration and feedback control based on embedded sensing.

A plausible implication is that the broader significance of MELEGROS lies less in a completed control architecture than in a materials-and-geometry paradigm. It advances the possibility that soft manipulators can be designed as single, sensorized, architected materials rather than as assemblies of body, actuator, and add-on sensor subsystems. Within the limits explicitly acknowledged, the work suggests a route toward larger continuum manipulators with embedded perception, graded stiffness, and more advanced closed-loop interaction.