A Millimeter-Scale Soft Robot for Tissue Biopsy Procedures

While interest in soft robotics as surgical tools has grown due to their inherently safe interactions with the body, their feasibility is limited in the amount of force that can be transmitted during procedures. This is especially apparent in minimally invasive procedures where millimeter-scale devices are necessary for reaching the desired surgical site, such as in interventional bronchoscopy. To leverage the benefits of soft robotics in minimally invasive surgery, a soft robot with integrated tip steering, stabilization, and needle deployment capabilities is proposed for lung tissue biopsy procedures. Design, fabrication, and modeling of the force transmission of this soft robotic platform allows for integration into a system with a diameter of 3.5 mm. Characterizations of the soft robot are performed to analyze bending angle, force transmission, and expansion during needle deployment. In-vitro experiments of both the needle deployment mechanism and fully integrated soft robot validate the proposed workflow and capabilities in a simulated surgical setting.

interactions, readily scaled designs, and more economical materials. [19][20][21] Safe robot-tissue interaction is a significant feature being leveraged for navigation of mounted MIS tools, such as needles and cameras. [22,23] Various soft continuum robots have been proposed for navigation through narrow, tortuous paths into regions of the body, such as the brain and lungs. [20] In these hard-to-reach anatomical areas, millimeterand submillimeter-scale robots have been demonstrated for their use as catheters that can deploy commonly used surgical tools, such as embolization coils, into the brain. [24,25] A magnetically actuated robotic catheter has demonstrated the ability to navigate through the vascular system of the brain to perform procedures with displayed benefits in dexterity and safety over current nonrobotic catheters. [26] For bronchoscopy, soft continuum robots have been scaled to reach the periphery of the lung. One such example uses a bending degree of freedom (DOF) and visionbased navigation to autonomously guide the robot in selected branches of the lung. [27] Other work plans a path ahead of the surgery based on preoperative imaging and uses magnetic actuation to achieve patient-specific navigation. [28] Out of the different actuation methods that have been used for surgical navigation, fluidic actuation brings many benefits such as minimized footprint of actuators in the deflated state, biocompatibility in the case of leakage, and noninterference with intraoperative imaging. [20,29,30] Furthermore, the integration of fluidic actuation into soft robots at the millimeter scale and smaller has demonstrated utility in the fabrication of increasingly complex actuators. [31] Previous works have leveraged these advantages in fluidic actuation to access further into the body; however, there remains a clinical need to address the issues in bronchoscopy of low diagnostic yields and variations in accuracy that arise due to factors such as breathing motion. Additionally, traditional surgical tools may be used with millimeter-scale, soft continuum robots; however, the produced forces of the robot scale with its cross-sectional area, [32] making them unable to interact with tissues that are stiffer than themselves. Finally, there are improvements required in the distal dexterity of current surgical tools to actively address the previously mentioned limitations in bronchoscopy. [9] Developed separately from navigational methods, soft robotic tools have been proposed for use in specific MIS procedures with the intention of increasing dexterity at the surgical site and reducing the physical strain of the surgeon. [33] Origami-inspired mechanisms present a promising addition to this class of tools with the unique ability to retain small form factors and fold/unfold into a desired orientation. [34][35][36] This design methodology usually entails the use of soft, hybrid materials that can be folded at defined compliant joints. [37] Origami actuators can utilize this compliance in tool-tissue interactions, such as brain retraction, where controlled, radial expansion has been shown to move the tissue with a safe level of force and simultaneously create defined working channels. [38] A Sarrus linkage robot for endoscopy demonstrates the ability of a foldable mechanism to deploy a needle for tissue biopsy by compressing the linkage once the capsule reaches the target area in the stomach. [39] Similarly, the combination of 2D layer-by-layer fabrication processes with micromanufactured, elastomeric actuators has resulted in surgical retraction tools that can be integrated with traditional endoscopes to provide more dexterity near the target area. [40,41] Inflatable actuators manufactured with soft films use similar principles [42] to create hybrid structures that can be deployed as tissue retraction devices [43] and as stabilization mechanisms [44] for traditional scopes. These works have enhanced the capabilities of surgeons in MIS; however, soft robotic end effectors, which provide the distal dexterity necessary in MIS procedures like biopsy, are yet to be integrated with navigational methods. This lack of integration is in part due to their need for greater force transmission when interacting with tissues at the millimeter scale, thus making these integrated soft robots susceptible to disturbances from their desired behaviors. [33] In this work, we propose a soft robotic platform that can steer the robot tip to reach a desired location in the lungs, actively increase tip force transmission via mechanical stabilization, and deploy a biopsy needle for tissue acquisition (Figure 1). These functionalities are all integrated within a 3.5 mm-diameter robot. This small diameter (almost half the size of traditional bronchoscopes) can potentially pave the way to access deeper peripheral locations within the lungs. The methods presented also lay the framework for further miniaturization in the future. We use an origami-inspired design methodology for controlled linear deployment of the needle from the tip of the soft, elastomeric body of the robot. The continuum body comprises a steering DOF for navigation to the target site and a stabilization DOF composed of a radially expansive, soft actuator to anchor the robot tip around the surrounding anatomy and counteract reaction forces during tissue puncturing. The resulting increase in force transmission is modeled using finite-element analysis (FEA) and the force output through the needle is modeled analytically to predict the expected force interactions with the target tissue. These predictions are validated through experimental

Design Overview
The soft robot integrated three fluidic actuated DOFs for steering, stabilization, and needle deployment. Steering and stabilization DOFs were embedded together in an elastomeric continuum body with an outer diameter of 3.5 mm. A radially expansive actuator acted as the stabilization mechanism embedded onto the tip of the continuum body. The mechanical stabilization provided by this actuator enabled the robot to anchor to the surrounding lung tissue as a means of increasing the force transmission of the needle. An origami-inspired, bellows actuator was integrated at the robot tip to achieve controlled, linear motion for deployment of a biopsy needle. The soft robot integrated all these capabilities to provide a fully soft platform with each DOF actuated independently ( Figure 1, Movie S1, Supporting Information).

Steering
A fluidic bending DOF was implemented to steer the tip of the soft continuum body when pressurized (Figure 2a). A 0.38 mmdiameter channel ran along the axis of the cylindrical body ( Figure 2b) creating the steering actuation chamber. The channel was located off-center on the circular cross section to promote a difference in strain along the axis and thus, creating the bending motion in a single direction. While it was primarily used for steering into the target lung branch, this bending DOF could also orient the needle deployment mechanism located at the tip of the robot. As shown in Figure 2b, the steering channel ran along the full length of the robot and was sealed at the tip. Pressurizing this channel independently from the stabilization mechanism enabled the robot to aim the needle at the tip after stabilization was deployed. This feature allows for the reorientation and aiming of the needle from the anchored position, if deemed necessary by the clinician during a procedure.

Stabilization
A radially expansive soft actuator, shown actuated in Figure 2a, was designed for integration into the continuum robot to safely anchor the tip in a desired position. Embedding the mechanism into the continuum body enables the creation of a monolithic design that achieves a final diameter, D SM , of 3.5 mm ( Figure 2b). The stabilization mechanism consisted of a fluidic chamber around the circumference of the robot located near its tip. The base of the chamber was embedded into the continuum body of the robot and, the top of the chamber comprised an inflation membrane made of Ecoflex 00-30 (Smooth-On, PA, USA), a material with a lesser durometer than the base (DragonSkin 10, Smooth-On, PA, USA), to create the radially expansive behavior (materials described in Section 2.2.3). Radial expansion was additionally promoted with the 0.7 mm inflation membrane thickness represented in Figure 2b as t m . Because of the limited space that the mechanism was designed to be operated within, t m was chosen to provide sufficient thickness that inflated at www.advancedsciencenews.com www.advintellsyst.com pressures below 50 kPa while also maintaining the final diameter, D SM , of the device. Figure 2b shows that t m encompasses the thickness of the inflation membrane and the thickness of the actuation chamber, which is a result of the fabrication method described in the next section.

Fabrication
The soft robot continuum body was fabricated through a series of molding processes shown in Figure 3a-c and Movie S2, Supporting Information. Two molds were used, each of which were used to cast a different material. The molds along with their end caps were all 3D printed using a FormLabs stereolithography (SLA) printer (Form 2, Formlabs Inc., MA, USA). The end caps for each mold were interchangeable to preserve alignment throughout the fabrication process. An added benefit of this feature is that, in the future, further robot diameter scaling is possible since the same end caps may be used with further miniaturized versions of the molds. The molding manufacturing technique relies on the resolution of the 3D-printed features in the mold; therefore, the continuum body can be readily miniaturized as the employed SLA printer has a resolution down to the micrometer scale. A three-part, cylindrical injection mold, shown as transparent in Figure 3a, was used first to create the base of the robot.
DragonSkin 10 Medium (Smooth-On, PA, USA) mixed with 10% by weight OS-2 solvent (DowSil, Dow Inc. MI, USA) was injected into this mold from one of the ends (Figure 3a). An end cap fit with alignment pins was fixed with mounting screws to this end. These alignment pins resulted in two 0.38 mm-diameter channels that ran along the length of the robot with the noncentral channel becoming the steering channel in a later step (Figure 2b). The open end of the mold was then injected until the mold was filled. It was placed with the open end upright and degassed at À100 kPa in a vacuum chamber for 3 min for 2 cycles. The final end cap was aligned with the pins and fixed to the open end of the mold before curing at 70°C for 20 min. This mold outlined the embedded actuation chamber and line that ran along the length of the robot, as shown in Figure 3b. To realize these features, a masking technique was used in which thin-film Teflon (PTFE) tape (Taega Technologies NC, USA) was wrapped around the robot masking the chamber. This technique ensured that the inflation membrane fabricated in subsequent steps sealed the chamber and left space for inflation of the radial actuator. The PTFE tape was laser cut using a 5 W diode-pumped solid state (DPSS) laser (Matrix-355, Coherent, Inc., CA, USA) to obtain the desired mask shape. The mask was then wrapped around the tip of the base at the outlined actuation chamber, as shown in Movie S2, Supporting Information. The alignment pins were reinserted into one end cap and the channels on the masked base. The second, four-part mold consisted of two halves which were each filled with Ecoflex 00-30 (Smooth-On, PA, USA), as shown in Figure 3c. The masked base was aligned to each of the halves using mounting screws and sealed leaving only the tip end open. This mold was degassed at À80 kPa for 3 min in the same open, upright configuration for 2 cycles. Finally, the final end cap was replaced to ensure the base was aligned within the mold and was cured using the same parameters. The resulting continuum body was completed by sealing the steering channel at the tip and inserting 0.635 mm tubing (MicroRenathane, Braintree Scientific, USA) into the radial stabilization actuator and steering channel.

Design
As shown in Figure 1, needle deployment was performed by an independent actuator that was fixed at the tip of the robot. The size of the actuator (3 mm in outer diameter) does not increase the outer diameter of the robot. Once the robot navigates to and stabilizes at the appropriate lesion, pneumatic actuation is used to deploy the needle for tissue biopsy.
The actuator was designed as a bellows structure to achieve linear expansion upon pressurization, as shown in Figure 2c. Circular actuation chambers that were made out of 0.038 mmthick thermoplastic elastomer (TPE) (Stretchlon 200, FiberGlast USA) were bonded on top of each other, in series, to create the bellows, as shown in Figure 2d. PTFE film was used as a masking material for each individual chamber. Previous work used a similar fabrication technology to demonstrate actuators with chamber diameters down to 9 mm; [43] however, we miniaturized the scale of our actuator to achieve a chamber diameter of 2.5 mm. With respect to previous work, our design miniaturized the bonding area, t b , to 0.25 mm, allowing each bellow chamber to achieve greater expansion at an overall diameter, D NDM , of 3 mm ( Figure 2d).
Stainless steel plates were designed to fix a needle-tubing assembly to the bellows actuator, as shown in Figure 2c. A hole centered on the plates had a diameter of 1.4 mm, which was 0.3 mm larger than the diameter of the bond connecting two bellows, allowing it to fit snugly between them. A 0.53 mm hole for holding the needle-tubing assembly on the top plate was located on the edge of D NDM . A similar hole in the bottom plate was 1 mm to allow the assembly to slide when the needle deployment mechanism was actuated. Aspiration of the biopsy sample could be performed from outside of the patient's body, as currently done in traditional interventional bronchoscopy. Therefore, the needle did not need to be removed or exchanged during a procedure and was designed to be fixed at the robot tip.

Fabrication
To fabricate the needle deployment actuator, a specific recipe for layering the thin TPE and PTFE films and applying heat and pressure was developed (Movie S2, Supporting Information). Prior to layering, TPE and PTFE films were laser cut with the DPSS laser into the previously outlined circular dimensions, as shown in Figure 3d. Similar to the fabrication of the continuum body, the needle deployment actuator may be scaled down further in the future as this technique provides a fabrication method in which the main features can be readily scaled. Miniaturization is possible due to the ability of the DPSS laser used in this method to cut features down to 15 μm. The recipes for heat and pressure may be modified to better suit a smaller scale.
These laser cut designs remained attached to a larger layer of film for alignment purposes, as shown in Figure 3e. All TPE and PTFE layers had a central channel through which each chamber received pressurization, except for the bottom and top TPE layers, which sealed the bellows into a closed system. The middle layers were placed in an alignment assembly using dowel pins. A pin through the center of the bellows aided in alignment of the inner PTFE layers, shown in Figure 3e, that masked each chamber. Each bellow was masked using 2.5 mm-diameter PTFE leaving t b to be 0.25 mm, as previously mentioned. A total of nine bellows were used in series in the final needle deployment mechanism. The stack of middle layers was heat pressed at 140°C under 140 kPa pressure for 1 min just to ensure that all layers maintained their aligned positions. After this, the central dowel pin was removed so the top and bottom TPE layers could be added onto the resulting laminate. This final laminate was heat pressed again under the same conditions for 7 min. The actuator was allowed to cool before being manually cut to release the actuator from the laminate (Figure 3f ).
Stainless steel plates of 0.076 mm thickness were also laser cut with the DPSS laser and secured between the top and bottom bellows to create the needle mounting plates, as shown in Figure 3f. The bottom plate was fixed between two bellows at the base of the actuator and the top plate was fixed between the top two bellows. The 0.635 mm tubing (MicroRenathane, Braintree Scientific, USA) was then inserted into the actuator through the inlet channel in the bottom bellow and sealed with Sil-Poxy (Smooth-On, PA, USA). A needle was fabricated by laser cutting a 30 gauge syringe needle with a 75°cut. The needle was attached to 0.635 mm tubing (MicroRenathane, Braintree Scientific, USA) to mimic the needles used to collect tissue samples for biopsy. This needle-tubing assembly was threaded through the holes in each stainless steel plate and the needle deployment mechanism was fixed to the tip of the continuum body using Sil-Poxy, as shown in Figure 3g. The tubing used for the needle deployment mechanism ran along the side of the robot and did not hinder any of its functionalities. The 0.635 mm-diameter tubing that was used may be integrated into the continuum body in the future using an axial channel similar to those used in the continuum body fabrication. The fabricated robot is shown actuating each DOF in Movie S1, Supporting Information.

Modeling
A finite-element model of the stabilization mechanism was developed using Abaqus (Simulia, Dassault Systems) FEA software to determine the expected performance of the device when placed in a surgical setting and design our device to be able to achieve the required range of forces. The inflation membrane is www.advancedsciencenews.com www.advintellsyst.com modeled based on the thickness, t m , of the device at the location of the actuation chamber. DragonSkin properties were applied to the base and Ecoflex properties were applied to the membrane using the Ogden hyperelastic model defined by parameters from Xavier et al. [45] Young's modulus, Poisson's ratio, and density of PTFE [46] were applied to the actuation chamber section on the base to model the wrapped PTFE mask. Lung tissue has also been modeled using hyperelastic models. [47] Therefore, parameters for the Mooney-Rivlin model for lung tissue [48] were applied to the anatomical channel shown in pink in Figure 4a,b. C3D10 hybrid elements were used to mesh both the stabilization mechanism and the channel using a seed size of 1.7 mm. The coarser mesh was applied to ensure convergence of the simulation and reduce computation time. [45] A pressure of 30 kPa was applied to the actuator chamber simulating the radial expansion of the membrane. An encastre boundary condition was applied to the inlet end of the stabilization mechanism, allowing it to act as a cantilever. The resulting shape of the membrane is shown in Figure 4d, where the membrane displaces radially by a maximum of 9.8 mm. The model assumes uniform inflation making it symmetric about its axis. However, as shown in Figure 4e, the fabricated device experiences nonuniform inflation when pressurized in free space. This discrepancy was found to be the result of a small overlap of the PTFE mask on itself, making the membrane thin enough in this region to experience greater inflation than other regions around the circumference of the stabilization mechanism. However, the stabilization mechanism is only intended to be deployed within an anatomical channel that is slightly larger in diameter as is modeled in our experimental setups (see Section 4.2 and Section 4.4). The size of the anatomical channel ensures that contact is made by the stabilization mechanism around the full circumference of the channel and as a result, the contact area around the channel is approximately equal, mirroring the predicted behavior in Figure 4c. A similar discrepancy, due to the actuation chamber being modeled as having space between the inflation membrane and mask, causes the model to inflate at slightly lower pressures than the fabricated device, leading to an offset of 8 kPa. Contact interaction between the mechanism and the anatomical channel was modeled using the penalty formulation with a 0.7 coefficient of friction based on measured trends from another study [49] for low-input force interactions of Ecoflex. Frictionless self-contact is also modeled for the inflation membrane of the mechanism. Inflation to 30 kPa was also simulated within the channel (Figure 4a) with a tip displacement of 2 mm applied in a subsequent step (Figure 4b). The 2 mm displacement is representative of the potential displacement that could be experienced by the robot tip during a biopsy procedure. The output of interest from this FEA model is the reaction forces at the tip, results shown in Figure 4c, which models the effective force transmission through the tip of the inflated stabilization mechanism. This data was exported from Abaqus to be plotted against experimental test data obtained using the same parameters (see Section 4.2). Reaction forces at the channel are also modeled indicating that the forces due to contact between the mechanism and channel are less than 0.03 N at any element node on the channel (Figure 4c). The reaction force at element nodes on one half of the channel surface is summed to also be plotted against test data (see Section 4.2). The model predicts that the force increases linearly with pressure after contacting the channel at 16 kPa, reaching 1.33 N in lateral forces at 30 kPa on one half of the anatomical channel. Average lateral forces in rigid bronchoscopy, which have caused tissue trauma in some patients, reach 9.43 N, [50] predicting the safety of our device at a level of force % 7Â less than this average. Furthermore, since our model demonstrates that this is a distributed force, the force at any given point on the channel is less than 0.03 N, further attributing to the safety of the robot when interacting with lung tissue. The total reaction force at the tip was shown to increase with linear behavior over the step that displacement was applied reaching a maximum of 0.67 N. Previously, it was determined that to puncture the soft tissue in the lungs, an average force of roughly 88 mN is required. [27] However, a tumor is roughly 2-3 times stiffer than average lung tissue, [51] meaning that the required force for biopsy may be closer to 264 mN. This model predicts b) The tip displacement is applied to the inflated mechanism, deforming it within the channel. c) The calculated reaction forces at the tip and on the channel are shown to be below 0.03 N at any node. The sum of the reaction forces at the tip represent the total force transmission through the tip at 2 mm displacement. d,e) A fabricated stabilization mechanism is shown outside of the anatomical channel and shows a similar behavior to the FEA model in terms of actuator expansion. that the addition of the stabilization mechanism to an anatomical channel will increase the puncture force of the robot beyond those required for bronchoscopy. The blocked force of the needle deployment mechanism can be determined by relating the force it produces directly to the cross-sectional area of the chamber and the pressure applied by F ¼ P Â A as in other studies. [43,44] The simple relationship approximates that the needle deployment mechanism at the proposed scale should be able to produce a maximum force of 1.22 N with an applied pressure of 250 kPa. This force is beyond the necessary forces for tissue biopsy which allows our design to account for any decrease in force when the needle deployment mechanism is allowed to expand toward the tissue. Conventional bronchoscopies have been known to use jet ventilation within the lungs at pressures up to 340 kPa, [52] which adequately covers the range of pressures used in the soft robot.

Bending Characterization
The bending DOF provided through pressurization of the steering channel was characterized by mounting the fully integrated soft robot horizontally so that bending occurred in the vertical direction, as shown in Figure 5a. The steering channel was then pressurized using a syringe pump (Harvard Apparatus Pump 11 Elite, USA) and pressure transducer (BSP000W, Balluff Inc. KY, USA) in 5 kPa steps up to 115 kPa. An image was captured using a DSLR camera (Nikon D7500, Japan) at each pressure step. Each image was imported into MATLAB (Mathworks, MA, USA) for image analysis of the bending angle.
The bending angle is plotted against pressure in Figure 5b showing the working range of the robot to be between 80 and 115 kPa. Between these pressures, the robot experiences its full range of bending. As shown in the plot (Figure 5b), the robot does not experience bending until the pressure reaches 80 kPa. The robot reaches a maximum bending angle of 196°a t 115 kPa, thus beyond full retroflexion (i.e., 180°).

Stabilization Characterization
Characterization of the stabilization mechanism was performed using an in vitro setup consisting of a mock anatomical channel made out of polycarbonate tubing with an inner diameter of 6 mm. The channel was mounted to a laser-cut acrylic platform and the continuum body of the robot (without the needle deployment mechanism) was fixed at its base to the acrylic platform, as shown in Figure 6a. To validate the design of the stabilization mechanism, the force transmission was assessed using two ATI Nano17 force sensors, one to detect the force transmitted through the tip during displacement and one to measure the contact force between the wall of the channel and the stabilization mechanism when inflated ( Figure 6a). Pressure was controlled using LabVIEW (v18.0, National Instruments, TX, USA) with an ITV0010 pressure regulator (SMC Corporation Tokyo, JP). The tip sensor was fit with a 3D-printed fixture to contact the tip of the continuum body and displace it downward by 2 mm at a speed of 2 mm min À1 by fixing it to an Instron (5943 Instron, USA). This was repeated for three instances when the stabilization mechanism was pressurized to 30 kPa and for three instances when it was not deployed. The results of these experiments are plotted in Figure 6b against the FEA model behavior (described in Section 3) with the average force represented by the solid line and the standard deviation represented by the shaded region. Over the course of the tip displacement test, contact force remained constant at an average of 1.28 N when the stabilization mechanism was deployed. The results of this experiment display an increase in force transmission over small displacements with the mechanism deployed. As the displacement of the tip increases, the force that is transmitted increases at a faster rate than without stabilization. This rate is defined as the effective stiffness of the robot in N mm À1 . The effective stiffness found from the experiments was 0.355N mm À1 , which is comparable to the modeled stiffness of 0.384N mm À1 . There is 7.5 % error between the model (please refer to Section 3) and the experimental behavior. The effective stiffness found without deployment of the stabilization mechanism was 0.063N mm À1 . The mechanical anchoring effect of the stabilization mechanism can be attributed to this behavior representing a 500% increase in the effective stiffness. Using stiffness as a measure of force transmission, the observed characterizations Our design also ensures that contact is made between the robot and the anatomical channel thus, ensuring that anchoring will be effectively achieved. This behavior is expected since inflation within the channel behaves as predicted by the model. As shown in Figure 6c, the mechanism begins to contact the 6 mm-diameter channel after being pressurized up to 20 kPa. As the pressure is increased, the contact force on the channel walls also increases with approximately linear behavior.
The results of the force transmission experiments show that operating the stabilization mechanism at 30 kPa allows us to maintain safe contact with the channels when compared to the lateral forces in conventional bronchoscopy, [50] while also enabling it to anchor to the anatomical channel. The relatively low operating pressure, with respect to 340 kPa that has been used in the lungs, [52] further guarantees safety in the case of leakage. The resulting contact force measurements, shown in Figure 6c, follow the same trend as the model with an offset of 8 kPa applied to the model to account for the actuation chamber discrepancy. An average error of 5.8% is calculated between the modeled contact force and experimental data at the same pressures, verifying that the effect of this discrepancy and the model assumption of uniform deformation mentioned in Section 3 does not introduce significant error. The resulting error demonstrates that the model sufficiently captures the observed behavior of the robot.
When contact is initiated with the channel walls, small displacement of the needle mounted at the tip may be experienced due to the location of the robot relative to the channel walls or initial nonuniform inflation. Displacement tracking tests were performed to quantify the effect of stabilization deployment on needle positioning. The fully integrated robot is inserted into a 3D-printed lung model that is also used later for in vitro experiments (see Section 4.4.2 for more details on lung model), as shown in Figure 7a. An Aurora Electromagnetic (E/M) Tracking System (Northern Digital, ON, CA) is used with a cylindrical sensor (0.8 mm diameter, 9 mm long) (Aurora Micro 6 DOF Sensor, Northern Digital, ON, CA) fixed in the needlemounting plate where the needle is intended to be. The stabilization mechanism is fully pressurized to reach contact with the surrounding anatomical walls (as illustrated in Figure 7a). The position of the needle within the plane of the anatomical channel is recorded during this timeframe as xand y-components of displacement (please refer to the reference frame in Figure 7a). The magnitude of displacement within the xy plane was computed and is plotted in Figure 7b showing displacements of 0.44 AE 0.04 mm. The effect due to stabilization is minimal, relative to the scale of average lung tumors (% 10mm). [9] The effect is also likely to be diminished because of the relative size of the anatomical channel. Furthermore, the ability of the robot to steer and adjust the tip position after inflation compensates for this potential effect, as demonstrated in Section 4.4.2.

Needle Deployment Characterization
A series of characterization tests were conducted to validate the needle deployment actuator's capability to produce blocked force and linear expansion.
The blocked force of the needle deployment actuator was measured using an ATI Nano17 sensor, as shown in Figure 8a. The actuator was mounted onto a fixed laser-cut acrylic plate using 3M 467MP double-sided adhesive. A 3D-printed tip of cylindrical diameter 4 mm was mounted to the ATI Nano17 sensor and lowered directly onto the actuator by attaching the Nano17 assembly to an Instron. The tip was mounted so that it was completely in Figure 6. a) Two ATI Nano17 sensors measure data on the tip force and contact forces during the displacement of the tip. b) The stiffness increase when the stabilization mechanism is deployed is measured as the slope of the force-displacement data. Measured data is plotted against the FEA model. c) Contact force versus pressure shows a linear increase in the force on the channel walls with force values similar to those found in the model. b,c) The solid line is the mean value and the shaded region is the standard deviation from three experiments.
www.advancedsciencenews.com www.advintellsyst.com contact with the top of the actuator, prior to the beginning of the test, and the actuator would experience no expansion when pressurized. The actuator was pressurized to a maximum of 220 kPa using a syringe pump (Harvard Apparatus Pump 11 Elite, USA), injecting at a flow rate of 0.5mL s À1 . Pressure was recorded using the Baluff pressure transducer. Force and pressure data were recorded at 100 Hz by NI (National Instruments) DAQ (USB-6210, National Instruments, TX, USA) and LabVIEW program from the two sensors. The resulting relationship between force and pressure is plotted in Figure 8b. As expected from our simplified force model, the force follows a linear relationship with pressure. At the maximum pressure, 220 kPa, the needle deployment actuator is able to produce %1.1 N on average. This is a similar result to the prediction at 220 kPa of 0.98 N, making the maximum error about 12% from the model. The force produced is beyond a force of 260 mN below pressures of 100 kPa, ensuring that the needle deployment mechanism will be able to puncture tissue wherever the tissue is within its range of expansion.
To characterize the stroke of the actuator, linear expansion was measured using the Aurora E/M Tracking System (Northern Digital, ON, CA). A fixed 3D-printed base was used for mounting the actuator inside a 6 mm tube to replicate the use case inside an anatomical channel (Figure 9a). A cylindrical tracking sensor (0.8 mm diameter, 9 mm long) (Aurora Micro 6 DOF Sensor, Northern Digital, ON, CA) was oriented at the tip of the actuator in a planar E/M field using 3M 9877 double-sided adhesive, as shown in Figure 9a. Upon inflation, the sensor was displaced by the actuator along the x-axis of the field, giving a direct measurement of expansion. The same syringe pump parameters from the blocked force experiments were used for pressurization. Displacement of the sensor was recorded over this period at 40 Hz sample rate and pressure was recorded using the same pressure transducer and NI DAQ.
The actuator follows a behavior, shown in Figure 9b, in which there is an initial rapid expansion over lower pressures until %30 kPa when the rate of expansion begins to decrease and the change in pressure produces a lesser level of expansion. The initial expansion of TPE is due to the geometry of the actuation chamber where expansion is quicker due to the small spaces masked by the PTFE layers between each bellow. At greater pressures than 30 kPa, expansion is determined by the strain of the TPE. In the deflated state, the bellows have an overall height of %1 mm. An average maximum expansion of 9.3 mm is achieved by the needle deployment actuator providing an %9:1 expansion ratio. This result is achieved at the 3 mm-diameter scale using only nine bellows as a result of the 0.25 mm bonding area. Electromagnetic navigation in bronchoscopy has allowed metal, commercial tools to reach within 5.7 mm of the targeted tumors in peripheral areas of the lung. [53] The scale of the  proposed soft robotic platform allows for potential navigation within a similar distance. The linear stroke of the needle deployment mechanism accounts for this 5 mm range for active control over biopsy.
In addition to the characterized expansion, the needle deployment actuator can be analyzed in terms of speed. The expansion data collected previously can be analyzed as speed using the values for expansion at given pressures with the known duration and sample rates of the tests. Assuming that the air behaves as an ideal gas, this same expansion pressure data can be used to determine the speed behavior of the needle deployment mechanism for any given syringe pump flow rate. Maximum attainable speeds for flow rates of 0.75, 1, and 1.4mL s À1 were all calculated and are recorded in Table 1. These values were confirmed by performing expansion tests again at each of the flow rates with measurements also shown in the table. Results allow for confirmation of the general directly proportional dependence of actuator speed on the flow rate of the driving fluid, further informing how the needle deployment mechanism may be controlled by the surgeon. To further analyze the speed with which the needle deployment mechanism could be actuated, deflation speeds were characterized by measuring the time over which the needle deployment mechanism would reach a steady level of contraction measured via the E/M Tracker. This was done for two speeds: atmospheric deflation, in which the system was deflated by opening it to atmospheric pressure, and vacuum deflation, in which a syringe was used to manually pull a vacuum on the system. It is important to note that the needle deployment actuator was able to contract fully to its original state when being deflated with vacuum while the atmospheric deflation left the actuator slightly expanded at %2.5 mm. Time was recorded using the known sample rate of the Aurora system until the actuator was no longer contracting. For deflation speeds, the needle deployment actuator was measured to deflate to atmospheric pressure over an average of 13 s while vacuum deflation took an average of 4.8 s to fully contract into its original state. Average deflation times were calculated over three prototypes each with three measurements.

Needle Deployment Experiments
The needle deployment mechanism's puncturing capability was demonstrated using a 10% by weight gelatin-water mixture (Knox Gelatin) as a lung tissue simulator. Past studies for robotic bronchoscopy applications have used similar mixtures of gelatin to simulate interaction with lung tissue. [54] Figure 10a shows the setup used for these in vitro experiments with the ATI Nano17 sensor recording the force required to puncture the simulator. A 3D-printed fixture was used to position the base of the needle deployment mechanism %6 mm from the simulator. The needle deployment mechanism and tissue simulator were both fixed to the acrylic setup so that the mechanism actuates perpendicular to gravity, as shown in Figure 10b, as would be the case when a patient lies down. Inflation pressure was applied to the mechanism using the same pressure regulator control system from the stabilization mechanism characterizations. Two pressurization methods were performed to determine if speed would effect the value of the recorded puncturing force. The pressure was first provided in steps from 5 kPa up to 100 kPa, while the second method, seen in Movie S1, Supporting Information, was conducted by pressurizing instantly from 0 to 100 kPa. The force required to puncture was consistent between the two tests with an average value of 0.06 N. The force to puncture ex vivo porcine lung tissue was found to be 0.088 N. [27] Therefore, the behavior during puncture of actual lung tissue is mirrored well through the use of the in vitro gelatin tissue simulator. When the needle deployment mechanism  made contact with the tissue simulator, needle puncture was confirmed by injecting water with pink dye, as shown in Figure 10c. The amount of time the needle deployment mechanism took to puncture using the instant pressurization method was measured to obtain a sampling rate at which biopsy can be performed. This time was measured to be on average 1.75 s. Time was measured from the deflated state until the mechanism reached the maximum force to puncture using the known NI DAQ (USB-6210, National Instruments, TX, USA) sampling rate. Combining these in vitro parameters with the vacuum deflation time of 4.8 s from the needle deployment expansion characterization, we determined an approximate operating frequency of 0.15 Hz for the mechanism within a mock surgical environment. With a refined control system setup, this operating frequency can be tuned to meet the needs of the surgeon.

Integrated Soft Robot Experiments
In vitro tests were performed to demonstrate the clinical suitability of the robot in a simulated surgical setting consisting of a simplified anatomical model of the bronchial tree. This model was 3D printed using Formlabs Elastic 50A resin and placed close to the gelatin tissue simulator, as shown in Figure 11 and Movie S1, Supporting Information. The model imitates the size of the bronchial tree at approximately the fourth generation of branches. [55,56] The model is lubricated before the experiment to mimic the moisture generally found in the lung airways. The continuum body of the robot is dyed green for greater visualization within the lung model. First, the robot was inserted into the lung model ( Figure 11a) and the bending DOF was used to navigate to the site of the tumor, represented by gelatin, by steering into the necessary branch, as shown in Figure 11b-d. The stabilization mechanism was then actuated to anchor the robot within the model (Figure 11e). Anchoring restricts any further advancement of the robot; however, the needle can be reoriented, if deemed necessary by the clinician, using the steering DOF again before taking a biopsy sample, as displayed in Figure 11f,g. The inclusion of this step in the workflow guarantees that the needle can preserve the desired positioning even if the position may be disturbed by the small movements induced during stabilization deployment. These movements may occur as a result of small levels of nonuniform expansion, the location of the robot relative to the walls of the channel, or the flexibility of the lung tissue. After reorienting the needle with the steering DOF, the needle deployment mechanism was pressurized to puncture the tissue simulator. Biopsy was confirmed visually by injecting pink dye into the simulator, as shown in Figure 11h. The same workflow would be followed if a tumor was found further from the lung branches where puncture of the lung wall would occur [54] to reach the tumor. The in vitro setup demonstrates the workflow and capabilities of the fully integrated soft robotic platform to navigate to the region of interest, stabilize itself by anchoring to the surrounding anatomy, and puncture tissue to perform a biopsy. Furthermore, the setup verifies that the robot can provide sufficient force transmission through its tip to puncture tissue and perform a biopsy.

Conclusion
To increase the surgical capabilities that may be embedded into a millimeter-scale, soft robot, we present this fully integrated soft robotic platform for tissue biopsy in bronchoscopy. The system embeds tip steering, stabilization, and needle deployment DOFs into a soft continuum robot with a final diameter of 3.5 mm. The combination of the three DOFs provides benefits over currently employed systems, such as manual traditional and ultrathin bronchoscopes as well as robotic systems, through the introduction of greater range of motion and flexibility, increased tip force transmission for a soft robot, and a distally actuated needle at the millimeter scale. While ultrathin bronchoscopes must compromise force transmission for flexibility in their tools, the proposed soft robot actively resists this compromise through the combination of stabilization and steering DOFs. The distal needle deployment mechanism initiates the puncturing of a lesion from the Figure 10. a) The in vitro test setup measuring the force required for the needle deployment mechanism to puncture the gelatin tissue simulator. b) The deflated and c) inflated needle deployment mechanism is mounted horizontal to the tissue simulator. Injected pink dye confirms that the mechanism was able to puncture the simulator when expanded from %6 mm.
www.advancedsciencenews.com www.advintellsyst.com robot tip. This functionality presents further benefits over commercial manual and robotic systems which rely on transmitting the puncturing force through the body of the bronchoscope from the surgeon outside of the patient's body to the target lesion. The robot has a diameter smaller than that of traditional bronchoscopes (i.e., % 6mm) [8] and can, therefore, potentially access deeper, peripheral locations into the lungs, where cancer can remain undetected. While the 3.5 mm diameter of the robot is comparable to current commercial robotic bronchoscopes, the fabrication methods employed allow for future miniaturization of the robot where it may further exploit the benefits of its embedded capabilities. At smaller scales, the integrated DOFs may have even greater advantages over commercial bronchoscopes and allow for deeper navigation into the lungs. In the future, further miniaturization will be possible as a result of the resolution of our robot manufacturing approach relying on the 3D printer and DPSS laser precision micromachining system. Characterizations of each independent mechanism demonstrate the feasibility of this system to be used in biopsy procedures. The robot is able to achieve retroflexion with a maximum bending angle of 196°. Bending angles of this range display the distal dexterity contribution this robot can have in bronchoscopy when compared to the bending range of conventional manual bronchoscopes at 120° [ 8] and robotic bronchoscopes, like the Auris Monarch, at 180°. [14] Both manual bronchoscopes and the Monarch incorporate a rigid segment near the tip for the integration of electronic components such as cameras; however, the Monarch is still able to achieve a bending angle equivalent to the Ion Endoluminal System that does not have this segment. Commercial systems also require biopsy tools to pass through their working channels which affect their overall flexibility. The proposed robot similarly has its bending segment integrated slightly off from the tip; however, the robot achieves a bending range beyond the capabilities of these commercial bronchoscopy systems and does not require rigid tools to pass through its body. Furthermore, the flexibility demonstrated in the soft robot is comparable to levels of bending that are sufficient for navigating the lung anatomy when compared to other soft bronchoscopic robots. [27,28] The stabilization mechanism, which actively anchors the soft robot to its surroundings, increases the effective stiffness of the system by 5 times. Over displacements less than 2 mm, the tip produces 0.75 N of force, which can be transmitted through the needle deployment mechanism. With these significant increases in force transmission at the millimeter scale, the soft robot can effectively biopsy tumors which are 2-3 times stiffer than previously tested healthy tissue. [27,51] The anchoring created by the stabilization mechanism can also pave the way toward breathing motion compensation while maintaining safe levels of contact force. Safety of the robot is validated through the level of forces that are produced on the channel walls and the operating pressures associated with stabilization. Data obtained from testing demonstrated contact forces of 1.28 N on half of the channel wall for an operating pressure of 30 kPa. The measured force is % 7.4Â less than the reported lateral forces in conventional bronchoscopy of 9.43 N, that may be imparted on the patient's mouth, vocal chords, or trachea. [50] The chance of trauma caused by these forces may be reduced as a result of the significant reduction in contact force by the proposed device. Furthermore, the measured force is distributed on the channel wall, as seen in the FEA model, demonstrating that the force experienced on the wall at a given point is actually 0.03 N or less. The operating pressure is significantly lower than ventilation pressures of up to 340 kPa that have been used in the lungs, [52] ensuring the robot does not contribute to the risk of pressure-induced trauma in the case of leakage.
The origami-inspired design of the needle deployment mechanism provides a method for controlled, linear deployment of the biopsy needle after anchoring. The current method employed in robotic biopsy procedures is to manually feed a needle through a scope working channel to puncture tissue thus, limiting sample collection to the abilities of the surgeon. Current robotic platforms are limited because manual control from surgeons is Figure 11. The fully integrated soft robot deployed in a mock surgical setting within a 3D-printed flexible model of the lung. The robot is a) inserted into the lung model and b) advanced to the branching point. c) The robot steers to d) navigate into a chosen lung branch (with the steering DOF), e) stabilizes at the target (using the stabilization DOF), f,g) reorients the needle (with the steering DOF), and h) deploys the needle for biopsy (with the needle deployment DOF). The gelatin tissue simulator is punctured at the end of the robot's path and injected with pink dye for confirmation.
www.advancedsciencenews.com www.advintellsyst.com performed from outside of the body and relies heavily on intraoperative imaging to ensure the needle punctures the correct tissue. [14] The proposed needle deployment mechanism introduces distal dexterity to the tip and enables the surgeon to control the amount of extension necessary to puncture the tissue. To biopsy tumors in the peripheral regions of the lung, smaller, more flexible needles are required to reach the site, resulting in the needle being more prone to unwanted bending and thus, unable to fully puncture some tumors. [57] Force characterizations of the needle deployment mechanism show that, with a maximum applied force of 1.1 N, it can provide the necessary force when puncturing. Our needle deployment approach enables biopsies within a range from 1 to 9.3 mm from the tip of the robot with a controllable operating frequency of 0.15 Hz based on our characterization parameters. For comparison, the human arm muscle can generally extend and contract within 200 ms. [58] In addition to the manual needle operation by the surgeon, resistance between the needle and working channel may keep the operating time for a full puncture on the order of 10 2 ms equivalent to an operating frequency within the range of 1-5 Hz. Soft actuators have been proven to operate within this range of frequencies. [59] Therefore, an independent control system will allow the needle deployment mechanism to operate at a frequency tuned for the surgeon while puncturing along a controlled path. Beyond bronchoscopy applications, other regions of the body could similarly benefit from the capabilities in this proposed platform. For example, transnasal neurosurgery requires a similar range of forces of 10-260 mN. [60] Future work will look into ex vivo tissue and in vivo animal experiments to validate our soft robotic platform for biopsies on biological tissue. Further, we will look into integrating the lateral external tubing for needle deployment within the robot body. In the future, during in vivo testing, the robot may be introduced into the lungs using a thin, flexible overtube of 3.5 mm inner diameter (i.e., covering the needle to prevent potential tissue scratching during navigation for safety). The overtube will be a medical-grade tubing custom designed for flexibility and minimum wall thickness. Similar techniques with overtubes are commonly used in interventional endoscopy. [43,44,61] Robotic bronchoscopy platforms, like the Monarch, also use this technique in the form of an outer sheath [13] or to embed sensing functionalities. [18] The overtube may be removed (simply by being pulled manually by the clinician) once the navigation to the target is completed to enable activation of the stabilization and needle deployment.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.