Fabrication of the slipknot

All experiments were conducted at room temperature (21 °C). The loading rate was set at 10 mm min^–1 during the slipknot-tying process. To ensure the reproducibility of the slipknots, we used a standardized fabrication method with a 3D-printed string-wrapping board. The designed grooves of the board conformed to the spatial configuration of the slipknot (Extended Data Fig. 1). String was wrapped along these grooves to produce standardized slipknots. The upper right side and lower right end of each slipknot were clamped, and the untightened slipknot was subjected to tension using a Zwick/Roell Z010 testing machine until a preset force F_tying was achieved. Following this procedure, standardized slipknots with F_tying were fabricated.

High-speed camera recordings

We used a high-speed camera (C321, Phantom) to capture the evolution of slipknot configurations during release from the front view (Fig. 2a(i–v)). The camera operated at 11,000 f.p.s. with a 10-μs electronic shutter speed using macro lenses (100MM F2.8 CA-Dreamer Macro ×2 lenses, Anhui Changgeng Optics Technology). It recorded images at 640 × 128 pixels with a 10-bit depth.

Micro-CT scanning

We used a micro-CT system (Xradia 610 Versa, Zeiss) with a voxel size of 1.3 × 1.3 × 1.3 µm³ to scan slipknots and to reconstruct their 3D structures based on five typical states. The micro-CT system was set to operate at 120 kV and 17.5 W. Each scan involved an exposure time of 2,500 ms per projection, during which the slipknots were securely positioned and examined for approximately 170 min. We fabricated an in situ tensile fixture (Supplementary Fig. 18) that fit in the sample chamber of the micro-CT scanner. This fixture enabled application of a preset tensile force to the slipknot, allowing it to be gradually opened to any desired state and maintaining it in a stable configuration throughout the scanning process. Micro-CT modelling and analyses were performed using Amira 3D software (v.2021.1, Thermo Fisher Scientific).

FEM of slipknots

FEM of the string slipknot was performed using Abaqus 2020 and the Abaqus/Explicit solver for computations. A model representing the slipknot was formulated using 8-node linear brick mesh elements with reduced integration and hourglass control (C3D8R). The material behaviour of the string was characterized by elastoplasticity. Elasticity was captured by using linear elasticity with an elastic modulus of 2,700 MPa and a Poisson’s ratio of 0.49. The plasticity was captured by using a plasticity model of combined hardening by fitting cyclic tensile test data (Supplementary Note 1 and Supplementary Fig. 19). The contact condition was modelled using a general contact-type interaction with a friction coefficient of 0.16 (Extended Data Fig. 6).

The simulation processes of slipknot tying, tightening and opening were visualized. First, the string was tied into a slipknot by implementing prescribed displacement sequences to the key points along the string. Second, the slipknot was tightened with a preset force. The pre-tension was maintained in one step and was unloaded afterwards. Third, the slipknot was opened by moving the free end away from the fixed end.

Mechanical test

We conducted a series of mechanical tests to explore the mechanical characteristics of slipknots, including consistency tests, parametric relationship tests and stability tests. All slipknots were made of fluorocarbon string (YGK) and each test contained five samples.

In the consistency test, 500 slipknots with single-knot loops were fabricated with a string diameter of d = 0.235 mm and a preset force Ftying = 7.500 N. In the parametric relationship test, we studied the influence of the number of knot loops, the preset force Ftying and the string diameter d on the open force Fpeak. First, we used string (d = 0.235 mm) to fabricate slipknots with different knot-loop numbers (single, double or triple) and different preset forces of Ftying (2.500 N, 5.000 N, 7.500 N, 10.000 N, 12.500 N, 15.000 N and 17.500 N). Next, with Ftying = 5.000 N, we used string of different diameters (0.148 mm, 0.165 mm, 0.235 mm, 0.285 mm and 0.333 mm) to fabricate slipknots with different knot-loop numbers (single, double or triple). In the stability test, the slipknots were tested after a time period (day 1, 2, 4, 8, 16, 24 and 32). The tested slipknots had a string diameter of d = 0.235 mm and different knot-loop numbers (single, double or triple).

All the slipknot samples were tested using a Zwick/Roell Z010 testing machine. The slipknots were opened at a speed of 50 mm min–1, and the Fpeak values were recorded. Finally, in the stability test, slipknots (d = 0.235 mm) with different knot-loop numbers (single, double or triple) were opened at different speeds (5 mm min–1, 10 mm min–1, 20 mm min–1, 40 mm min–1, 60 mm min–1, 80 mm min–1 and 100 mm min–1), and the Fpeak values were recorded.

Friction test

We conducted friction tests to measure the kinetic friction coefficients of three types of filaments: fluorocarbon monofilament (d = 0.235 mm, YGK), braided absorbable sutures (4-0 Vicryl, Ethicon) and braided non-absorbable sutures (4-0 Mersilk, Ethicon). Tests were carried out under dry, moist and lubricated conditions. The moist conditions included synthetic blood (Phygene Biotechnology), simulated body fluid (Phygene Biotechnology) and natural saline (Phygene Biotechnology). The lubricated condition was silicone oil 10 mPa·s (Aladdin). We established a friction measurement fixture, whereby a filament was suspended above a spiral tube through a pulley system (Extended Data Fig. 3a). Bearings applied the controllable normal force (Fn) to press the filament firmly against the spiral tube, which was wrapped with the same filament. The spiral tube was placed in a tank to enable tests in moist and lubricated conditions. By steadily pulling the filament using a universal testing machine, the tangential force (Fτ) was measured.

Fτ under varying Fn was recorded and analysed through linear regression to determine the kinetic friction coefficient of the filament. The friction coefficients in moist and lubricated conditions were evaluated by dripping different volumes of liquids onto the filament–tube contact region. We also replaced the spiral tube with a silicone block to measure friction coefficients between the filament and silicone (Extended Data Fig. 3b–d).

Opening sliputures under dry, moist and lubricated conditions

To examine whether exposure to fluid in a real surgical environment interacts with the slipknot, which would further influence the Fpeak, we conducted a series of opening sliputure tests using various types of filaments, including fluorocarbon monofilament (YGK), braided absorbable sutures (Vicryl, Ethicon) and braided non-absorbable sutures (Mersilk, Ethicon) (Extended Data Fig. 4). Each material was tested under dry, moist and lubricated conditions, with five samples per group. For fluorocarbon monofilaments, we used string (d = 0.235 mm) to fabricate slipknots with double-knot loops and different preset forces of Ftying (2.500 N, 5.000 N, 7.500 N and 10.000 N). For braided absorbable sutures and non-absorbable sutures, strings (4-0, d = 0.150 mm) were used to fabricate slipknots with double-knot loops and different preset forces of Ftying (0.100 N, 0.200 N, 0.350 N, 0.400 N, 0.800 N, 1.600 N and 3.200 N).

All the samples were tested using a Zwick/Roell Z020 testing machine in dry conditions, in synthetic blood (Phygene Biotechnology), in simulated body fluid (Phygene Biotechnology), in natural saline (Phygene Biotechnology) and in silicone oil 10 mPa·s (Aladdin). The slipknots were released at a speed of 30 mm min–1, and the Fpeak values were recorded.

Wearable knot-tying force test platform

A wearable knot-tying force test platform (Supplementary Fig. 20) was constructed to accurately measure the tension in sutures during a surgical suturing process. The platform consisted of a knob, two pulleys, a slider, a steel wire and a load cell (SBT641C, Simbatouch). One end of the suture was secured to the slider through the knob, and the adjacent pulley guided the suture, thereby minimizing the impact of hand-posture variations on force measurement. The slider was connected to the load cell by a steel wire (d = 0.8 mm) running through the upper pulley. During suturing, the surgeon positioned their index finger through a hole in the device, with the thumb resting on the shell above the pulley next to the load cell. Tension in the suture was directly measured by the load cell as the suture was pulled through the platform.

Bimanual real-time dynamometer

A bimanual real-time dynamometer platform was developed to accurately capture tensile forces applied by both hands during surgical knot tying. The system consisted of a medical sterile interface, a passive revolute joint and a quick connector (Supplementary Fig. 21a). The sterile interface ensured clinical compatibility, and the passive revolute joint provided a free rotational range of 80° (Supplementary Fig. 21b), which enabled natural wrist movement and reduced the influence of hand posture or angle on measurements. To eliminate angular interference and to achieve high-fidelity force data, the dynamometer adopted a decoupled mechanical structure (Supplementary Fig. 21c), and the force applied through the suture was transmitted directly through the joint to the internal sensor. Notably, the platform supports simultaneous use of two devices, one for each hand, which enabled real-time, bilateral force measurement.

Knot-tying force test across surgeons

To assess the clinical feasibility of the sliputure transmitting force, we designed a knotting-force test platform to measure knotting-force precision and incision pressure (Supplementary Fig. 4). This platform comprises three major components: (1) the desired sliputure with a slipknot on a braided common suture (4-0 Mersilk, Ethicon); (2) a silicone practice model with pressure-sensitive films (4LW, Fuji film Prescale); and (3) a real-time dynamometer. Fabricating the sliputure with an Fpeak of 1.400 N (Ftying = 0.350 N), we evaluated the impact of sliputure transmitting force on the knotting-force performance of surgeons31,32. Ten junior surgeons and five senior surgeons were randomly selected and recruited for the experiment. Junior surgeons were defined as surgeons with fewer than 10 years of clinical experience, whereas senior surgeons were those with more than 10 years of clinical experience. Each surgeon performed knot tying (n = 20) with common sutures (n = 10) and sliputures (n = 10).

The real-time dynamometer collected data on the force used, and then we acquired the force data after data transformation (Supplementary Figs. 3a and 4). All force data were used for further comparative analyses.

Incision pressure test of standing sutures

To evaluate the mechanical transmission ability of the sliputure, we used a silicone practice model along with pressure-sensitive film (4LW, Fuji film Prescale) to quantify incision pressures of standing sutures using sliputures and common sutures (Supplementary Fig. 3b). In the silicone practice model, we performed a 7-cm incision by laser cutting to simulate a wound. Adjacent to the incision at 0.5 cm from either side, 2 sets of circular holes measuring 1 mm in diameter were introduced parallel to the trajectory of the incision. These circular holes facilitated the entry and exit of the needle during the suturing process. To eliminate any interference among needles, a longitudinal spacing of 1 cm was maintained between neighbouring circular holes. Leveraging the midpoint of the feasible force range, we prepared an engineered slipknot (4-0 Mersilk, Ethicon) with an Fpeak value of 1.300 N for subsequent experiments.

Pressure-sensitive films (film A for colour rendering and film B for pressure assessment) were precisely positioned to match the contour of the simulated wound. Following the process of suturing and pre-tying knots, the pressure exerted by the slipknot on both sides of the wound made a red colour in film B. To ensure consistency, senior surgeons performed the same suturing procedures using both sliputures (n = 25) and common sutures (n = 25) on the silicone practice board. These suturing trials were conducted under open, laparoscopic (Hefei DVL Electron) and robotic (Intuitive Surgical-Fosun Medical Technology) surgical settings. During robotic surgery, the knotting velocity of the surgeon was quantified using Tracker (v.6.0.9). Subsequently, the colour intensity of the pressure-sensitive film A, which corresponded to the applied pressure, was quantified by scanning the images and converting them into numerical values using commercial software (Fuji film FPD8010E v.2.5.0.3).

Incision pressure test of continuous sutures

To evaluate the stress distribution of continuous sutures, we performed incision pressure tests of continuous sutures³³ (n = 5) of five stitches on a silicone practice model (Extended Data Fig. 7 and Supplementary Note 4). After excluding the influence of the anchor knot (first stitch) and the final knot (last stitch) on incision pressure, we chose the middle three stitches based on the oblique trajectory of the suture in continuous sutures. Subsequently, the pressure-sensitive films (4LW, Fuji film Prescale) embedded in the silicone were removed, and the films corresponding to the middle three stitches were selected for analyses. Colour changes were quantified by scanning the images and converting them into numerical values using commercial software (Fuji film FPD8010E).

Bimanual knot-tying force test

The bimanual knot-tying force test was conducted to measure and compare the forces on both sides of the sliputures (Supplementary Fig. 22). Five junior surgeons and five senior surgeons were randomly recruited for the experiment. Each surgeon wore real-time dynamometers bimanually and performed knot tying using sliputures of double-knot loops (4-0 Mersilk, Ethicon) with an F_peak value of 1.300 N (n = 10). During the test, the side with the slipknot was pulled and the peak force was defined as T_active, whereas the other side without the slipknot was held stationary and the peak force was referred to as T_fixed. The process was stopped after the release of the slipknot. The F_peak values from both dynamometers were recorded for comparison.

The buffer zone of sliputures

The buffer zone of sliputures was determined through uniaxial tensile testing. Specifically, sliputures of double-knot loops (4-0 Mersilk, Ethicon) with an F_peak value of 1.300 N were used. Tests were conducted using a Zwick/Roell Z020 testing machine, equipped with an Xforce HP load cell (capacity of 200 N). Under uniaxial tensile loading of 10 mm min^–1, the sliputures released and then formed a buffer zone after relaxation, where the suture remained kinked and bore zero load (Extended Data Fig. 8).

Sliputures for continuous sutures

The sliputures used for continuous suture of five stitches was as follows. At the first stitch, we used a sliputure (Suture A, 4-0 Vicryl, Ethicon) to tie an anchor knot at the proximal end of the incision. Continuous sutures of four stitches were then performed using the same suture procedure as usual, leaving the final stitch unknotted. Subsequently, a second sliputure with two slipknots (Suture B, 4-0 Mersilk, Ethicon) was used to place one stitch at the distal end of the incision, with the slipknot near the incision being first used to form an anchor knot. Finally, two sutures were tied together to complete the final knot (Supplementary Fig. 12).

Feasible force-range exploration

To systematically explore the feasible force range for colonic injury repair, a mechanical test platform was assembled (Supplementary Fig. 5a,b). This platform included the specimens to be tested, a dedicated tensiometer (DS2-50N, Zhiqu), a digitally controlled electric machine, a pump, a manometer and PBS solution containing methylene blue. The experimental procedure (Supplementary Fig. 5c) involved the introduction of a 2-mm-diameter injury into an ex vivo rat colon, specifically those of Sprague–Dawley rats weighing between 200 and 250 g (n = 20). The injury was generated using a biopsy punch (Dynarex). Subsequently, the punctured colon was bathed in a continuous solution of methylene blue maintained at a pressure of 20 mmHg. Pre-suturing was carried out using common sutures (4-0 Mersilk, Ethicon). During the experimental phase, one end of the common suture remained secured, whereas the other end was horizontally pulled by the tensiometer at a controlled rate of 10 mm min–1.

The tension applied was consistently monitored in real-time throughout the process. Post-test analyses involved categorizing the tension as Fmin once the injury was successfully repaired without leakage. Conversely, the tension was labelled as Fmax when the colonic wall tore and leakage occurred. The boundaries of the feasible force range were subsequently determined based on the Fmin and Fmax values. The feasible force range can also be predicted by mechanical modelling (Supplementary Note 5).

Ex vivo and in vivo rat colonic injury repair

To compare the sliputure transmitting force in both ex vivo and in vivo conditions, we conducted experiments using female Sprague–Dawley rats weighing between 200 and 250 g. The rats were randomly divided into two groups: the ex vivo group (n = 5) and the in vivo group (n = 5). The experiments involving the sliputure (target Fpeak = 1.300 N) were carried out on an automated mechanical test platform (Supplementary Fig. 5a,b). For both ex vivo and in vivo tests, a 2-mm-diameter injury was introduced to the resected colon using a biopsy punch (Dynarex). Subsequently, the injury was repaired by the tensiometer at a consistent rate of 10 mm min–1 using a sliputure under 20 mmHg of intraluminal pressure. After the availability of the opened signal of the slipknot, we recorded the force data from the tensiometer and halted the electric machine to stop the tensiometer.

Afterwards, a reverse knot was tied to secure the surgical knots (surgeon’s knot). For the in vivo test, the distal colon and small intestine were clamped 2-cm apart from the repaired region. PBS solution containing methylene blue was introduced through the repair area to assess colonic leakage. In summary, we compared the force data and operation efficacy of colonic leakage between ex vivo and in vivo groups to evaluate the mechanical transmission of sliputures.

Comparison of anastomotic leakage, postsurgical adhesion and tissue ischaemia

Female Sprague–Dawley rats weighing between 200 and 250 g were randomly allocated into two groups: the sliputure group and the common suture group. This division aimed to facilitate a comparative study of anastomotic leakage, postsurgical adhesion and tissue ischaemia. The experimental procedures were performed by a junior surgeon with 2 years of clinical experience. This surgeon was affiliated with the Department of General Surgery of another tertiary hospital and had no conflict of interest. To minimize bias, the surgeon was blinded to the study objectives while tying the common sutures. Following a 24-h fasting period, rats were anaesthetized through intraperitoneal administration of ketamine (80 mg kg–1). The rats were then placed on a heated pad for the surgery. After abdominal hair was removed, laparotomy was performed to expose the colon. Using a biopsy punch (Dynarex), a 2-mm-diameter injury was created in the colon.

In the subgroup focusing on leakage, the injury was repaired using sliputures (n = 10) with an Fpeak value of 1.300 N, whereas common sutures were used in the control group (n = 10). To observe and compare colonic leakage, the distal colon and small intestine were clamped 2 cm away from the repaired region. Subsequently, PBS solution containing methylene blue was injected through the repair site. In the subgroup assessing adhesion, after colonic injury repair as described above, continuous sutures (4-0 Mersilk, Ethicon) were used to close the peritoneum, and interrupted sutures (4-0 Mersilk, Ethicon) were used to close the abdomen. On the seventh day postoperatively, the repaired colon was exposed and adhesion rates were observed and compared between the sliputure-treated group (n = 10) and the common-suture-treated group (n = 10).

In the subgroup assessing tissue ischaemia, visualization of vascular conditions before and after biopsy puncture, as well as after repair using sliputures or common sutures, were carried out. This visualization was accomplished by LSCI. The same procedures for colonic injury repair and abdomen closure were performed. After concluding of all tests, the rats were humanely euthanized by CO2 inhalation.

Comparison of wound healing

Female Sprague–Dawley rats weighing between 200 and 250 g were randomly allocated into two groups: the sliputure group and the common suture group. In each group, subgroups were divided as sliputure D1–D8 and common suture D1–D8 based on postsurgical days. The entire experimental process was carried out in a sterile environment. The preoperative preparations were consistent with the procedures outlined above. Through laparotomy, the colon of the rat was exposed, and a 2-mm-diameter injury was introduced using a biopsy punch (Dynarex). Sliputures and common sutures were used to repair the injury in the sliputure group and the common suture group, respectively. Postoperative assessments spanned from day 1 to day 8, which included animal euthanasia under CO2 inhalation, daily tissue collection and burst pressure measurement in both groups (n = 3).

On postoperative day 5, blood collection and analyses were conducted on a subset of rats in both the sliputure and common suture groups before humane euthanasia using CO2 inhalation. The regions after repair were excised, fixed in 10% formalin for 24 h and prepared for histological analyses.

LSCI

Vessel visualization of the targeted colon area was conducted before and after biopsy puncture, as well as following the repair procedures using sliputures or common sutures, using LSCI. To mitigate potential motion artefacts, the targeted colon region was gently externalized and positioned on gauzes for stabilization. The maintenance of a stable blood flow was achieved through the continuous use of 37 °C saline solution. Once the parameters were set, HD mode was used to capture a clear speckle image depicting the distribution of blood flow in the intact target region of the colon. The exposure time for image acquisition was set at 5 s. Subsequently, after the introduction of a 2-mm-diameter biopsy puncture, additional speckle images were obtained using LSCI under the same parameter settings and measurement time.

Following this, surgeons proceeded to repair the colonic injury with both sliputures and common sutures, and LSCI was used to capture speckle images with the same parameter settings and measurement time. This imaging process aimed to track and visualize the vascular dynamics of the targeted colon region before, during and after the surgical interventions to provide insights into the effects of the repair methods on blood flow and tissue perfusion. The data were collected using a Laser Speckle Contrast Imaging system RFLSI III (v.4.0).

H&E and multiplex immunofluorescence staining

Laparoscopic application of sliputures in a live porcine model

In vivo porcine experiments used female Bama miniature pigs (23–27 kg, 5–6 months of age) to demonstrate the application and function of the slipknot in laparoscopic surgery. The laparoscopic platform (Fig. 3j) was acquired from Karl Storz. After fasting for 24 h, pigs were premedicated with intramuscular atropine (0.04 mg kg–1), ketamine (10 mg kg–1) and midazolam (0.6 mg kg–1), induced with isoflurane (4%), endotracheally intubated and maintained with isoflurane (1–2%) during surgery. We placed the pigs in dorsal recumbency and aseptically prepared the abdominal region. Several trocars (Endopath Xcel, Ethicon and Kangji Medical) were placed on the abdomen to enable access of surgical instruments and a camera. Surgeons identified the colon and made a 1-cm-wide incision in the colon wall. After cleaning the lumen and disinfecting the intestinal incision, the intestinal injury was repaired using sliputures with inverting interrupted sutures (4-0 Mersilk, Ethicon).

A control group in which the injury in the colon wall was repaired using common sutures with inverting interrupted sutures (4-0 Mersilk, Ethicon) was also performed for comparison. Observations for intestinal leakage were made 10 min after the process. The abdominal wall was closed in a multilayer fashion, and the animal was euthanized after the experiment.

Robotic application of sliputures in a live porcine model

In vivo porcine experiments used female Bama miniature pigs (23–27 kg, 5–6 months of age) to demonstrate the application and function of the slipknot in robotic surgery. The robotic platform (Fig. 4a) was acquired from Intuitive Surgical-Fosun Medical Technology. After fasting for 24 h, pigs were premedicated with intramuscular atropine (0.04 mg kg–1), ketamine (10 mg kg–1) and midazolam (0.6 mg kg–1), induced with isoflurane (4%), endotracheally intubated and maintained with isoflurane (1–2%) during surgery. We placed the pigs in dorsal recumbency and aseptically prepared the abdominal region. Several trocars were placed on the abdomen to enable access of surgical instruments and a camera, and one additional trocar (Endopath Xcel, Ethicon) was placed to facilitate assistant operation. Surgeons identified the colon and made a 1-cm-wide incision in the colon wall. After cleaning the lumen and disinfecting the intestinal incision, the intestinal injury was repaired using sliputures with inverting interrupted sutures (4-0 Mersilk, Ethicon).

The control group underwent the same procedure, but the repair was made using common sutures. Observation for intestinal leakage were performed 10 min after the process. The abdominal wall was closed in a multilayer fashion, and the animal was euthanized after the experiment.

We performed additional live porcine experiments, including induction of acute colonic perforation with peritonitis and colonic injury repair of female Bama miniature pigs (23–27 kg, 5–6 months of age) to compare the effects of standing sutures using sliputures, staples and continuous sutures under a robotic platform (Intuitive Surgical-Fosun Medical Technology). After fasting for 24 h, pigs were anaesthetized following the same protocol as described above. We placed the pigs in dorsal recumbency and aseptically prepared the abdominal region. Several trocars were placed on the abdomen to enable access of surgical instruments and a camera. Surgeons identified the colon and made three 1-cm-wide incisions in the colon wall at intervals of 3 cm. The abdominal wall was closed in a multilayer fashion.

After fasting for 24 h, the porcine models of acute colonic perforation with peritonitis were anaesthetized following the same protocol as described above, and trocars were inserted at the same incision sites, with one additional trocar (Endopath Xcel, Ethicon) placed to facilitate assistant operation. Surgeons identified the colonic injury, disinfected the intestinal incisions and repaired them using continuous sutures, staples or standing sutures with sliputures. The appearance of tissue as the repaired injury site was recorded. Subsequently, indocyanine green (0.1 mg kg^–1) was intravenously administered, followed by a 5 ml saline flush. Under fluorescence imaging of the robotic system, colonic vascular perfusion was observed 60 s after injection. Finally, the abdominal cavity was irrigated with saline and the abdominal wall was closed in a multilayer fashion.

Development and validation of a vision-based sliputure and slipknot detection system

We used the da Vinci Xi robotic system enhanced by a purely vision-based sliputure and slipknot detection system to validate the practicality and applicability of our proposed sliputure as a surgical consumable and non-electronic haptic sensor. To model sliputure and slipknot detection, we formulated it as a regression modelling problem by decoupling this process into sliputure line extraction and opened slipknot detection given by the following equations: Lines = Extractor (I) and Open slipknot = Detector(Lines), where I denotes RGB images of the first-person view from the da Vinci Xi endoscope, and Lines denote a set of sliputure lines extracted by the function Extractor(·). The extracted sliputures Lines were then input into a slipknot detector function Detector(·), which predicts whether there is an opened slipknot.

Notably, the reason why we decoupled the modelling process into two parts is to fully leverage the advantages of data-driven methods by converting a specific medical problem into a general modelling problem while ensuring real-time performance for the surgical robotic system under limited data and instrument resources.

Specifically, for sliputure line extraction, we designed an encoder–decoder neural network based on a U-Net34 structure (named SlipknotNet, shown in Fig. 4a and Extended Data Fig. 9) to model the function Extractor(·), in which the image encoder is realized by a pretrained ResNet50 (ref. 35) and the decoder contains up-convolution operators, SENet36 modules and a suture-detection module. During the training stage, the parameters of the encoder were frozen using pretrained parameters from ImageNet37, whereas the parameters of the decoder were trained on over 1,800 annotated images by volunteer surgeons using our custom-made annotation software. We used a joint loss function including dice loss38 and focal loss39 using the following equation: Loss = α × dice loss + (1 – α) × focal loss. For the open slipknot detection, the function Detector(·) was realized by implementing template matching using the scale-invariant feature transform operator from OpenCV.

The entire process of sliputure extraction and slipknot detection is visualized in Fig. 4a.

In both silicone practice models and live porcine colonic injury repair using sliputures teleoperated by the da Vinci Xi robotic system under real-time monitoring in 30 Hz with a local workstation (Ubuntu 22.04, Intel Core i9-14900, NVIDIA RTX 4090m), the aforementioned sliputure and slipknot detection scheme ran with a circle mark. A green mark flashing means no open slipknot detected, which allows continued control of the robotic arm. A red mark means the slipknot is open, and a stop control signal is sent to the robot console by UART communication, and a high-level internal stop from the robot is triggered. Our engineering code and dataset are both open-source and available on our project website⁴⁰.

Automated system for standardized and rapid mass production of sliputures

The automated sliputure construction system comprises modules for wire feeding, force control, wire wrapping, slipknot collection and a user interface (Supplementary Fig. 16). The wire-feeding module supplies filament material (diameter of 0.1–2 mm, unlimited length) from a spool into wire-wrapping module, where it is precisely cut to the desired length. The wire-wrapping module uses coordinated motor-driven and pneumatic actuators to form slipknot configurations with the desired number of knot loops. The force control module, integrated with piezoelectric sensors and precision screw actuators, applies and maintains a F_tying (0.1–20 N) on the constructed slipknot. After setting up the parameters, the system operates autonomously without manual intervention, thereby ensuring standardized and rapid mass production of sliputures.

Vision-based robotic dual-arm suture system for the slipknot application

The vision-based robotic dual-arm suture system consisted of two Franka Emika Panda 7 degree-of-freedom arms and a camera. The overall framework of the vision-based dual-arm suture system is illustrated in Supplementary Fig. 14. The reference velocity of the active arm was set as 0.5 cm s^–1. Through the real-time processing of greyscale images, the active arm automatically stops based on the opened signal of the slipknot (Supplementary Fig. 15).

DNA slipknot molecular dynamic simulation

Coarse-grained molecular dynamic simulations were performed using the program Gromacs with the Martini 2 force field for nucleic acids. The initial straight DNA structure with randomly assigned sequences had a length of approximately 670 nm and a diameter of 2.2 nm, comprising 130,000 atoms and 28,000 beads. A DNA knot was artificially constructed in the middle of the straight DNA. A triclinic simulation box with periodic boundaries in all three directions was used (Supplementary Fig. 17).

The atomic potential energy was minimized first using 1,000 static steps. In the dynamic simulations, we used a time step of 10 fs. The pressure was maintained at 1 bar with the Nosé–Hoover Langevin piston and the temperature was maintained at 273 K using a Langevin thermostat. The DNA knot was untangled by pulling one end of the DNA at a velocity of 0.01 nm ps^–1. The total simulation time for untangling the DNA knot was around 5 ns. The atomic structure was visualized using visual molecular dynamics.

Slipknot-enhanced safe human–robot interactions

Building on the demonstrated clinical success and the inherent capability of slipknots to precisely limit force release, we extended its versatility to robotics by integrating sliputures into a custom-built 4 degree-of-freedom cable-driven robotic arm, engineered as an adjustable mechanical fuse to ensure safe human–robot interactions (Fig. 4g,h, Extended Data Fig. 10 and Supplementary Video 16). The actuation architecture incorporated two direct-drive systems for the base and the shoulder, as well as two independent tendon routes that drive the bidirectional motion of the elbow and wrist joints, with a slipknot strategically embedded in the forearm segment to monitor and regulate inter-joint forces. The actuation transmission pathway is instantaneously interrupted when the tension force at the sliputure exceeds a pre-programmed threshold Fpeak due to an overloaded external interaction, achieved through slipknot-mediated cable elongation that dissipates stored mechanical energy.

This enables simultaneous protection of humans and delicate objects from excessive force and the safeguarding of robotic components from structural overload during impact events. It is worth noting that the capability to flexibly adjust the Fpeak of the slipknot enables on-demand control of contact loads, a critical feature for human–robot interactions. Unlike traditional solutions, this purely mechanical safety solution operates without requiring auxiliary electronics or additional mass at the end-effector, demonstrating its applicability for seamless integration into existing robotic platforms.

Statistical analysis

Categorical variables were described using frequency and percentage, which were assessed using Fisher’s exact tests or Chi-square tests between groups. Continuous variables are presented as the mean and standard deviations, which were compared using Wilcoxon rank-sum tests or Student’s t-tests. GraphPad Prism (v.9.5.0, GraphPad Software) was used for all statistical analyses in the study. In statistical analyses, the significance thresholds were considered as *P < 0.05, **P ≤ 0.01 and ***P ≤ 0.001.

Animal studies

Study design

No formal sample size calculation was performed, and the number of samples was determined by the maximum available resources. In animal experiments that explored the feasible force range, a sample size of n = 20 was used. For ex vivo and in vivo rat colonic injury repair, the sample sizes were n = 5 for ex vivo and n = 5 for in vivo experiments. For comparison of anastomotic complications, the sample sizes were n = 10 in the sliputure group and n = 10 in the control group. For comparison of wound healing, the sample sizes were n = 3 for each subgroup from days 1–8 in both the sliputure and control groups. Laparoscopic and robotic experiments were conducted with a minimum of one pig per group. No statistical methods were applied to calculate sample sizes. The animal samples prepared by the same method were randomly allocated to each group. Blinding was not used in this study.

Animals

The following animals were used in experiments: female, specific pathogen-free Sprague–Dawley rats aged 4 weeks and weighing 200−250 g; and female Bama miniature pigs aged 5–6 months and weighing 23−27 kg.

Ethics

All animal studies were approved by the Institutional Animal Care and Use Committee at Zhejiang University (ZJU20230084 and ZJU20250067), and postoperative care was supervised at the Animal Experimental Center of Sir Run-Run Shaw Hospital, Zhejiang University.

Surgeon participant studies

Informed consent was obtained from all participants, and the study was approved by the Institutional Review Board of Sir Run Run Shaw Hospital, Zhejiang University (approval numbers 2023-0528 and 2025-0023).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.