Shaping the Future: Flexible Robots with Tendon Control

by

Author: Denis Avetisyan

Researchers present a complete design, fabrication, and modeling pipeline for creating highly accurate, tendon-driven continuum robots with tapered, flexible polymer spines.

The design features a tapered, flexible backbone of thermoplastic polyurethane-a continuum robot enabled by integrated electronics and the precise actuation of three equally spaced tendons-demonstrating an approach to robotic morphology where adaptability stems from material properties and distributed control.
The design features a tapered, flexible backbone of thermoplastic polyurethane-a continuum robot enabled by integrated electronics and the precise actuation of three equally spaced tendons-demonstrating an approach to robotic morphology where adaptability stems from material properties and distributed control.

This work details a framework leveraging parametric CAD, kinetostatic modeling, and 3D printing to achieve centimeter-level shape prediction in soft robots with tapered TPU backbones.

While many continuum robots sacrifice adaptability for simplicity or incur high costs for specialized functionality, this work-Tendon-Actuated Robots with a Tapered, Flexible Polymer Backbone: Design, Fabrication, and Modeling-introduces a customizable and low-cost framework for 3D-printed, tendon-driven robots featuring tapered backbones. By extending Cosserat rod theory with a Newtonian approach to account for spatially varying geometry, we achieve centimeter-level accuracy in shape prediction through a combination of parametric CAD automation, kinetostatic modeling, and experimental validation. This enables enhanced distal compliance and broadens the scope of potential applications, from robotic inspection to delicate manipulation-but how can this design be further optimized for specific, complex tasks in unstructured environments?

The Allure of Adaptability: Embracing Continuum Robotics

Continuum robots, distinguished by their infinitely-flexible bodies, present a compelling solution for navigating the complexities of constrained environments – spaces too tight or irregular for conventional rigid robots. However, this very flexibility introduces a significant engineering hurdle: predictable curvature. Unlike robots composed of discrete joints, where movement is precisely controlled, the bending of a continuum robot is governed by a complex interplay of material properties and external forces. Achieving a desired shape isn’t simply a matter of issuing commands; it requires overcoming the challenges of modeling and controlling an infinite number of degrees of freedom. This makes precise manipulation and reliable operation – crucial for tasks like minimally-invasive surgery or search and rescue – exceedingly difficult, necessitating innovative approaches to both design and control algorithms to unlock the full potential of these adaptable machines.

Conventional robotics relies on modeling movement through discrete, rigid links – a methodology fundamentally ill-suited to the dynamics of continuum robots. These machines, designed to mimic the flexibility of biological organisms, present an infinite number of potential configurations, defying the simplified assumptions inherent in rigid-body calculations. Attempting to predict the bending, twisting, and overall deformation of a continuously flexible structure using traditional methods results in significant inaccuracies and control challenges. The very nature of continuum robots-their lack of discrete joints-means that established kinematic and dynamic equations break down, requiring researchers to develop entirely new mathematical frameworks and computational techniques capable of capturing the complex interplay between material properties, external forces, and the robot’s resulting curvature. This shift demands a move away from treating the robot as a series of connected solids and towards understanding it as a continuous, deformable body governed by principles of elasticity and continuum mechanics.

Achieving both strength and dexterity in continuum robots hinges on innovative backbone designs and materials science. Traditional robotics relies on rigid links, but these structures are ill-suited for the infinite degrees of freedom inherent in flexible robots. Researchers are exploring materials like fiber-reinforced polymers and shape memory alloys, alongside designs inspired by natural structures such as elephant trunks or octopus arms. These approaches prioritize distributing stress along the robot’s length, preventing buckling or failure during bending and extension. Furthermore, variable stiffness materials – those capable of changing rigidity on demand – offer a promising pathway to dynamically adjust maneuverability and load-bearing capacity, allowing for precise control and robust operation in challenging environments. The ultimate goal is to create a backbone that is simultaneously resilient enough to withstand external forces and flexible enough to navigate complex obstacles.

Effective grasping with continuum robots isn’t simply about reaching an object; it demands precise, predictable curvature, and research indicates a logarithmic spiral represents an ideal form for many manipulation tasks. This isn’t arbitrary – the logarithmic spiral, frequently observed in natural systems like nautilus shells and sunflower seed arrangements, maximizes stability and force distribution during contact. Achieving this shape allows the robot to envelop an object securely, distributing grasping forces evenly to prevent slippage or damage. Furthermore, the spiral’s self-similar geometry offers robustness to slight errors in positioning or material properties, meaning even imperfectly formed curves can still maintain a functional grasp. Consequently, engineers are increasingly focused on designing actuators and control algorithms capable of reliably generating and maintaining this specific curled conformation, bringing bio-inspired robots closer to the dexterity seen in natural appendages.

A 3D-printed, tendon-actuated continuum robot is shown with Vicon motion capture markers attached to its outer surface for tracking and analysis.
A 3D-printed, tendon-actuated continuum robot is shown with Vicon motion capture markers attached to its outer surface for tracking and analysis.

A Kinetostatic Framework: Modeling the Continuous Form

Cosserat Rod Theory, utilized within this model, describes the mechanics of deformable bodies by representing them as continuous rods with both bending and torsional stiffness, differing from Euler-Bernoulli beam theory which assumes plane sections remain planar. This approach accounts for the effects of shear deformation and allows for the independent specification of bending and twisting rigidities, crucial for accurately modeling slender, flexible structures undergoing complex deformations. The theory mathematically defines the kinematics and statics of these rods through [latex]\textbf{v} = \textbf{x}’ + \boldsymbol{\omega} \times \textbf{x}[/latex], where [latex]\textbf{v}[/latex] is the velocity field, [latex]\textbf{x}'[/latex] is the material derivative of the position vector, and [latex]\boldsymbol{\omega}[/latex] represents the rotation vector, enabling precise calculations of deformation and internal stresses within the robot’s flexible components.

Tendon actuation within the robot’s design provides precise control over bending and extension movements. This is achieved by strategically routing tendons along the robot’s structure; when these tendons are actuated – that is, when tension is applied – they induce curvature due to the resulting moments. The degree of bending or extension is directly proportional to the applied tendon force and inversely proportional to the robot’s [latex]EI[/latex] (flexural rigidity). Multiple tendons are utilized to enable control over multiple degrees of freedom and complex trajectories, allowing for nuanced and repeatable movements. This method offers a lightweight and efficient means of actuation compared to traditional methods like motors and linkages, particularly well-suited for slender, flexible robots.

Accurate implementation of the kinetostatic model relies on precise material property values, specifically Young’s Modulus. This parameter was determined through a systematic line search, evaluating values between 50 MPa and 200 MPa with 1 MPa resolution. This iterative calibration process identified the Young’s Modulus that best corresponds to the observed physical behavior of the flexible structure, ensuring the model’s predictive capabilities are optimized for the robot’s specific construction and materials. The selected value directly influences the calculation of deflection and stress within the structure, impacting the accuracy of curvature prediction.

The assumption of inextensible tendons is a key simplification within the kinetostatic model, significantly reducing computational demands without substantial compromise to overall accuracy. While real tendons exhibit some degree of elasticity, treating them as perfectly rigid streamlines the kinematic and dynamic calculations, eliminating the need to model complex material deformation. This simplification allows for faster simulations and real-time control implementations. The impact on model fidelity is minimized because the tendons primarily transmit tensile forces, and their contribution to structural deformation is comparatively small in the overall bending and extension behavior of the slender robot. This approach provides a pragmatic balance between model complexity and computational efficiency, facilitating practical application of the kinetostatic model.

The Cosserat rod model defines a coordinate frame transformation [latex]g(s)[/latex] mapping points along the robot's backbone arc length [latex]s[/latex] from the proximal, mounted frame [latex]{x,y,z}[/latex] to the local cross-sectional frame [latex]{x^{b},y^{b},z^{b}}[/latex].
The Cosserat rod model defines a coordinate frame transformation [latex]g(s)[/latex] mapping points along the robot’s backbone arc length [latex]s[/latex] from the proximal, mounted frame [latex]{x,y,z}[/latex] to the local cross-sectional frame [latex]{x^{b},y^{b},z^{b}}[/latex].

Validation Through Physical Realization: From Simulation to Embodiment

The robot’s backbone was designed using Parametric CAD software, enabling iterative adjustments to its geometry. This approach facilitated the creation of a tapered design, where the cross-sectional dimensions vary along the length of the structure. Optimization focused on maximizing curvature while maintaining structural integrity. By defining key dimensions as parameters, we could systematically explore a range of geometries and identify configurations that best met performance criteria. The tapered design, achieved through parametric control, directly contributes to the robot’s ability to achieve the desired range of motion and adapt to varying terrain.

The robot backbone was fabricated from Thermoplastic Polyurethane (TPU) due to its inherent flexibility and compatibility with additive manufacturing processes. TPU is a class of polyurethane with high elasticity, allowing for repeated bending and flexing without permanent deformation, which is crucial for a dynamically curving structure. Its suitability for 3D printing enables the creation of complex geometries directly, minimizing assembly and maximizing design freedom. The material properties of TPU were selected to balance flexibility with structural integrity, ensuring the backbone could maintain its shape under load while facilitating the desired curvature profiles during robotic locomotion.

Vicon motion capture technology was implemented to acquire high-precision 3D positional data of the robot, achieving a resolution of 0.02 mm. This system utilizes infrared cameras to track reflective markers placed on the robot, allowing for real-time determination of its kinematic state. The resulting data served as ground truth for validating the kinetostatic model, enabling quantitative comparison between predicted and actual robot pose. The high resolution of the Vicon system was critical for accurately capturing subtle deformations of the robot’s tapered backbone and ensuring the fidelity of the validation process.

Data from load cells, boasting a resolution of 0.125 N/bit, were integrated as input parameters into the robot’s kinetostatic model. This high-resolution force data allowed for refinement of the curvature prediction algorithm, resulting in experimental validation demonstrating centimeter-level accuracy between the predicted and physically measured backbone shapes. The integration of load cell data significantly improved the fidelity of the model and its ability to accurately represent the robot’s physical behavior under load.

Simulations demonstrate that backbone shape is significantly influenced by both taper angle and cable tension, with a [latex]67 MPa[/latex] Young’s modulus TPU material exhibiting predictable deformation over a [latex]34.5 cm[/latex] length and [latex]1.11 cm[/latex] base radius.
Simulations demonstrate that backbone shape is significantly influenced by both taper angle and cable tension, with a [latex]67 MPa[/latex] Young’s modulus TPU material exhibiting predictable deformation over a [latex]34.5 cm[/latex] length and [latex]1.11 cm[/latex] base radius.

Toward Adaptive Systems: Implications and Future Trajectories

This research establishes a robust framework for the design of continuum robots specifically suited for operation within challenging, constrained spaces. Unlike traditional robots relying on rigid links, these robots-inspired by biological organisms like octopuses and elephant trunks-deform continuously, allowing them to navigate narrow passages, squeeze through openings, and conform to irregular surfaces. The presented approach moves beyond simply building such robots; it provides a systematic methodology for predicting and optimizing their performance within specific environments. By accurately modeling the interplay between robot geometry, material properties, and external constraints, designers can proactively address issues of maneuverability, stability, and load-bearing capacity, ultimately leading to more effective and versatile robots for applications ranging from minimally invasive surgery to search and rescue operations in disaster zones.

A significant advantage of this research lies in its ability to efficiently address the inverse design problem for continuum robots. Traditionally, determining the necessary geometric parameters to achieve a desired shape or force distribution has been computationally expensive. This model, however, streamlines the process, allowing researchers to quickly explore a vast design space and identify optimal configurations for specific tasks. By drastically reducing the time required for design iteration, the work accelerates the development of specialized robots tailored to navigate challenging environments – from minimally invasive surgical tools to adaptable manipulators for space exploration. This efficiency isn’t merely incremental; it represents a shift from painstakingly crafted designs to a more agile and responsive development cycle, ultimately fostering innovation in the field of soft robotics.

Compared to computationally intensive methods like the Finite Element Method, this research demonstrates a kinetostatic approach offers a significantly more efficient solution for modeling and designing continuum robots. Traditional simulations often require extensive processing power and time to accurately represent the complex deformations of these robots within constrained spaces. However, by focusing on the static equilibrium and force distribution – the ‘kinetostatics’ – the model drastically reduces computational demands without sacrificing accuracy in predicting robot behavior. This efficiency is particularly crucial for iterative design processes and real-time control applications, enabling rapid prototyping and deployment of specialized continuum robots tailored for intricate environments. The speed and scalability of this kinetostatic method position it as a valuable tool for researchers and engineers seeking to advance the field of soft robotics.

Continued development anticipates integrating more nuanced material properties into the robot design process, moving beyond simplified assumptions to accurately reflect the behavior of real-world materials like elastomers and composites. This will necessitate refinements to the current kinetostatic model to capture phenomena such as viscoelasticity and hysteresis, ultimately yielding more precise simulations and improved robot performance. Simultaneously, research will extend to advanced control methodologies, including reinforcement learning and model predictive control, to enable these robots to autonomously navigate highly complex and unpredictable environments. Such strategies will allow for real-time adaptation to unforeseen obstacles and optimization of movement for tasks requiring dexterity and precision, paving the way for applications in minimally invasive surgery, search and rescue, and in-space exploration.

The pursuit of increasingly complex robotic systems, as demonstrated by this work on tendon-actuated robots with tapered backbones, inevitably introduces vulnerabilities. Each refinement, each layer of automation in parametric CAD and kinetostatic modeling, represents a potential point of failure – a step towards maturity through the acceptance of imperfection. As Bertrand Russell observed, “The good life is one inspired by love and guided by knowledge.” This sentiment echoes the design philosophy inherent in continuum robotics; the ‘knowledge’ gained through meticulous modeling and validation is applied to ‘inspire’ more resilient and adaptable systems, acknowledging that even the most elegant design will require iterative correction and refinement over time. The centimeter-level accuracy achieved isn’t a destination, but a marker along a continuous path of improvement.

What’s Next?

The pursuit of increasingly accurate kinematic prediction in continuum robotics is, at its core, a prolonged negotiation with entropy. This work demonstrates a notable refinement in shape prediction – centimeter-level accuracy is not negligible – yet it also highlights the inherent limitations of any model attempting to capture the infinite degrees of freedom within a flexible structure. Versioning this design – iterative refinements of the tapered backbone and actuation scheme – is a form of memory, acknowledging that each iteration is built upon the accumulated imperfections of its predecessors.

The arrow of time always points toward refactoring. Future efforts will inevitably address the modeling of hysteresis, creep, and the long-term effects of cyclic loading on the TPU material. More intriguing, however, is the potential to move beyond purely kinematic control. Integrating sensing directly into the polymer structure – creating a robot that knows its own deformation – promises a shift from open-loop prediction to closed-loop adaptation.

Ultimately, the success of this approach – and indeed, the field of soft robotics as a whole – will be measured not by the fidelity of its models, but by its resilience. A truly graceful aging process for these systems requires not just accurate prediction, but the capacity to self-correct, to learn from deviation, and to maintain functionality even as the inevitable decay sets in.

Original article: https://arxiv.org/pdf/2603.19124.pdf

Contact the author: https://www.linkedin.com/in/avetisyan/

See also:

2026-03-22 14:20