Design of a soft robotic endoscope with enhanced bending and AI-based prediction

Gentler Tools for Keyhole Surgery

Many modern surgeries are now done through tiny cuts or natural openings rather than large incisions. This approach, called minimally invasive surgery, helps patients recover faster with less pain. But to make it work well, doctors need long, snakelike tools that can steer through the body’s winding pathways without harming delicate tissues. This paper introduces a new kind of soft, flexible endoscope that bends more easily with lower internal pressure, along with a smart computer model that can predict how it will move. Together, they point toward surgical tools that are both safer for patients and easier for surgeons to control.

Why Traditional Scopes Fall Short

Conventional endoscopes sit at two extremes. Rigid ones are strong and precise but struggle to navigate tight curves deep inside the body. Very flexible ones can reach farther but may buckle, loop, or lose control, making it harder for surgeons to work accurately. Soft robotic endoscopes try to combine the best of both worlds by using rubberlike materials that bend smoothly when their internal chambers are pressurized. Still, many current designs need relatively high pressure to achieve useful bending, which raises safety concerns: if pressure inside the device approaches or exceeds blood pressure, a failure could damage tissue. Earlier designs also suffered from friction, ballooning, or unpredictable motion, and they often required heavy computer simulations every time the design changed.

A New Bending Body

The researchers propose a redesigned soft endoscope shaped like a short cylinder with five internal air chambers arranged around its center. Each chamber has a half-moon cross section rather than a simple circle, which helps generate stronger bending with less unwanted bulging. A light outer fiber sheath wraps the device, preventing it from ballooning outward when air is pumped in and keeping its shape stable. By carefully choosing the number, shape, and placement of the chambers, the team created a module that can reach a bending angle of about 90 degrees using only 0.2 bar of pressure—lower than typical blood pressure and more than 10 percent better than a leading previous design under similar conditions. This makes it easier to steer around sharp corners while keeping loads on the surrounding tissues low.

Testing the Design in the Virtual World

To understand exactly how their endoscope would behave, the authors first relied on detailed computer simulations. Using a method called finite element analysis, they modeled the soft silicone body and its outer fiber layer while air pressures inside the chambers were varied. This allowed them to predict how much the device would bend or twist and to check whether the material would remain safely within its limits. The team experimented with different mesh resolutions—how finely the device is broken up for calculation—until the results no longer changed significantly, a sign that the predictions were stable. These simulations showed that adjusting which chambers are pressurized, and by how much, can produce controlled bending in different directions and, if desired, twisting around the central axis.

Letting Artificial Intelligence Learn the Motion

Running such detailed simulations for every new pressure pattern or design tweak would be slow and computationally costly. To overcome this, the researchers built a data-driven prediction model using machine learning. They generated hundreds of example cases in Python, each describing a unique combination of pressures in the five chambers and the resulting bending and twisting motions. Two types of learning algorithms—artificial neural networks and support vector machines—were trained to map from pressures to four outcomes: bending angle, twisting angle, and how quickly each of these motions occurs. After cleaning and scaling the data, the team trained and tested these models, finding that both could reproduce the simulation results closely, with the neural network performing slightly better overall.

From Simulation to Real Hardware

To confirm that their virtual models matched reality, the authors built physical molds with a 3D printer and cast the endoscope body in a very stretchy silicone. They then pressurized the chambers in carefully controlled steps and measured how much the real device bent. The experimental bending angles closely followed the simulation predictions, with small differences explained by natural variations in the silicone and the simplifications in the mathematical model. Repeating the same test multiple times produced nearly identical bending angles, showing that the device behaves reliably. Across a range of pressures, the prototype achieved higher bending than previous designs at safer, lower pressures, validating both the mechanical concept and the AI-based prediction tools.

What This Means for Patients and Surgeons

In everyday terms, this work brings us closer to surgical tools that move more like a gentle fingertip than a rigid stick. The new soft endoscope can bend sharply while staying within safe pressure limits, helping surgeons reach hidden or hard-to-access areas with less risk of injury. The accompanying AI model acts like a fast "lookup table" that tells designers and, in the future, control systems how the scope will respond before it moves, without needing to rerun heavy simulations each time. With further development—such as linking multiple segments, adding real-time sensing, and closing the loop between measurements and control—this combination of soft robotics and artificial intelligence could make minimally invasive procedures safer, more precise, and available to more patients.

Citation: Hani, M., Elghitany, M.N., Sweif, R. et al. Design of a soft robotic endoscope with enhanced bending and AI-based prediction. Sci Rep 16, 12878 (2026). https://doi.org/10.1038/s41598-026-46334-y

Keywords: soft robotic endoscope, minimally invasive surgery, pneumatic soft actuators, medical robotics, artificial intelligence modeling