How Continuum Robots Handle Friction When They Press Against the World

How Continuum Robots Handle Friction When They Press Against the World

Jia Shen, Junhyoung Ha, Yue Chen

7 min read•Nov 16, 2026

A new model helps flexible continuum robots account for friction when they touch walls and other surroundings—a contact behavior that frictionless models can miss. In a wall-contact test, it reduced the robot’s maximum tip-position error from 45.4 mm to 5.85 mm. The approach represents contact and friction as a linear complementarity problem, linking the robot’s motion to whether its tip sticks or slides.

What Did the Researchers Build?

Jia Shen, Junhyoung Ha, and Yue Chen developed a model for predicting how a continuum robot moves while rubbing against its environment. Continuum robots are flexible machines—such as steerable catheters or slender inspection tools—that bend along their length rather than moving through a small number of rigid joints. Their flexibility lets them enter confined spaces, but it also makes their contact with walls difficult to predict.

The model focuses on external friction: friction between the robot and its surroundings. That differs from internal friction, such as rubbing between the nested tubes of a concentric tube robot. Earlier continuum-robot friction models addressed internal effects or assumed that sliding friction was already at its maximum. The new approach instead represents contact constraints and friction as part of the robot’s motion calculation, so it can account for sticking as well as sliding.

The researchers first set out a general formulation, then worked through a specific elastic-rod example using the Cosserat rod framework, a way to describe a flexible rod’s bending and twisting. They also tested the approach on a planar tendon-driven continuum robot interacting with a rigid wall. The aim is not simply to detect that contact exists, but to predict how contact forces change the robot’s shape and trajectory.

Illustration of contact types considered for a flexible robot

What Were the Key Results?

The clearest quantitative result comes from an experiment with a 287.9 mm-long tendon-driven continuum robot pressing against an aluminum wall. The friction-aware model’s maximum tip-position error was 5.85 mm, compared with 45.4 mm for the frictionless model. Its maximum bending-angle error was 4.6°, compared with 60.6° without friction. These reported maximum errors correspond to reductions of about 87% and 92%, respectively.

The experiment also showed why friction matters: the frictionless model did not capture the tip’s sticking motion. The friction-aware model’s largest errors occurred near the transition where the robot lost contact with the wall, indicating that predicting changes in contact state remains challenging.

MeasurementFriction-aware modelFrictionless model
Maximum tip-position error5.85 mm45.4 mm
Maximum bending-angle error4.6°60.6°
Robot length287.9 mm287.9 mm

The friction coefficient used in the experiment was 0.092. The team obtained it by minimizing the mean trajectory error, so it was a fitted value for this test rather than an independently reported material measurement. The paper also develops an elastic-rod case and discusses simulated contact scenarios, but the supplied results do not give numerical benchmark comparisons for those cases.

Flowchart of the contact and friction modeling process

How Does the Model Work?

The central tool is a linear complementarity problem, or LCP. An LCP expresses a system using variables that must be nonnegative and conditions that must hold in complementary pairs. In contact mechanics, this structure can capture the idea that a wall cannot pull a robot toward itself: either the robot is separated from the wall, or a contact force acts to prevent penetration. The model also uses contact constraints, abbreviated as CCs, to connect those contact conditions to the robot’s motion.

A standard LCP can be written as \(w = Mz + q\), with \(z \geq 0\), \(w \geq 0\), and \(z^\mathsf{T}w = 0\). Here, the variables encode quantities such as contact forces and constraint violations; the complementary condition means that a force and its corresponding violation cannot both be active. The paper builds a contact-and-friction formulation around this mathematical structure. Rather than assuming the contact force in advance, it solves for contact conditions together with the robot’s motion.

That coupling matters for friction. When a tip sticks, its movement along the surface is constrained; when it slides, friction opposes that motion. The model’s state therefore depends on how the robot moves and on the contact situation detected along the way. The paper notes that this requires an intricate contact-detection process. In the elastic-rod case, the researchers derive the contact-related matrices needed to instantiate the general formulation. They also show that, when friction is removed, their formulation matches frictionless contact kinematics from earlier work.

The same modeling approach is then applied to a planar tendon-driven continuum robot. Tendons actuate an outer catheter and a notched needle, while base translation allows axial compression. This provides a concrete test of whether the contact model can describe a flexible robot’s changing shape as it interacts with a rigid surface.

Simulation example of a continuum robot contacting a plane

Why Does This Matter for Robotics?

For flexible robots, contact is often part of the task rather than an exception. A catheter, inspection probe, or other slender manipulator may slide along a surface, press against an obstacle, or temporarily stick while advancing through a narrow route. If a model assumes frictionless contact, it can predict the wrong tip position and robot shape—as the experiment’s large error difference demonstrates.

More accurate contact prediction can support better motion planning, simulation, and control. A planner that accounts for sticking and sliding has a stronger basis for estimating where a flexible robot will move when its surroundings constrain it. That is useful wherever the robot must navigate close to walls or other structures. The experiment is a modeling validation, however; it does not by itself demonstrate a complete autonomous system or prove improved performance across every application.

The result also highlights a practical engineering point: robot behavior depends on the interaction between a machine and its environment, not just on its actuators. Teams evaluating robot hardware can compare options through Robot Overflow’s used robots for sale; this research addresses one part of the engineering challenge that arises after a flexible robot is deployed—predicting how it behaves in contact.

What Are the Limitations and Open Questions?

The physical validation covers one planar robot interacting with one rigid aluminum wall. Its fitted friction coefficient of 0.092 was selected to reduce trajectory error, so further testing would be needed to establish how reliably the model transfers across materials, surface finishes, robot designs, and operating conditions.

The largest remaining errors occurred around the point where wall contact was lost. The researchers point to unmodeled stick-slip motion and internal friction from the tendons as possible causes. The contact-detection process is also described as intricate, which could complicate implementation in larger or more complex environments. The reported comparison is promising, but the provided results do not establish computation time, robustness across varied tests, or performance in three-dimensional contact.

What Questions Do Engineers Ask?

What problem does the model address? It predicts how external friction changes a continuum robot’s motion when the robot contacts its surroundings, including cases where the tip sticks rather than slides.

What is a linear complementarity problem? It is a mathematical framework for conditions that activate in complementary pairs, such as a non-penetration constraint and its contact force.

How much did the model improve the wall-contact results? The maximum tip-position error fell from 45.4 mm to 5.85 mm, while maximum bending-angle error fell from 60.6° to 4.6°.

Does the experiment prove the model works for every continuum robot? No. It validates the approach on a particular planar, tendon-driven robot and rigid wall; broader testing is still needed.

Conclusion

The study shows that representing friction as part of a continuum robot’s contact and motion calculation can substantially improve trajectory prediction in a wall-contact test. The next challenge is establishing how well the method performs across more robots, surfaces, and contact transitions.