---
kind: concept
slug: tendon-driven-actuation
title: Tendon-driven actuation
summary: >-
  Moving a flexible or articulated part by pulling a tendon, cable or backbone from an actuator placed away from the moving part. The actuator can sit outside the instrument, which keeps the instrument small. The trade-off: friction, backlash and stretch in the line make the position and the tip force hard to predict and to sense.
tags: [robotics, actuation, surgical-robotics, medical-robotics]
papers: ["1807.03731"]
related: [continuum-robot, magnetic-actuation, piecewise-constant-curvature]
status: draft
updated_at: 2026-11-12
---

## Overview

In tendon-driven actuation, a line runs from a motor or actuator to the part
it moves. It is called a tendon, a wire, a cable or, for a continuum arm, a
backbone. Pulling the line bends or moves the part at the far end. The review
describes the general reason for this design: the actuators are placed
remotely from the end effector, which helps miniaturise the instrument
[arxiv:1807.03731 §3.3]. Robots designed for natural orifice and minimally invasive
surgery are mostly wire-actuated, with recent growth in continuum robots as the
exception [arxiv:1807.03731 §3.3].

The cost sits in the line itself. The review states that compliance, motion
losses and friction in the actuation lines reduce the accuracy of the end
effector's motion. They also weaken force sensing and can reduce control
stability [arxiv:1807.03731 §3.3]. A model that maps motor angle to tip
position is therefore biased, and the bias depends on load and history. The
rest of this page covers the design variants the review names, the losses it
lists, and the compensation methods it reports.

## Variants

### Wire-actuated articulated end effector

Serial arms with a jointed wrist are driven by wires that run from the base to
the joints. The review cites the da Vinci single-site system and the SAIT-KAIST
single-port system as examples. Both use articulated serial structures that
rely on wire actuation to control their end effectors [arxiv:1807.03731 §3.1,
fig. 4]. The joints give dexterity at the tool tip, but the wires add the
friction and stretch described above [arxiv:1807.03731 §3.3]. Card:
[Medical Technologies and Challenges of Robot Assisted Minimally Invasive
Intervention and Diagnostics](/papers/1807.03731).

### Multi-backbone push-pull continuum arm

A multi-backbone continuum architecture drives several backbones in push-pull
pairs. The review gives the IREP robot as an example. It uses push-pull
actuation on a number of superelastic NiTi backbones to set the bending shape
of each segment [arxiv:1807.03731 §3.1, fig. 4]. Friction and coupling models
have been built for multi-backbone robots of this kind [arxiv:1807.03731 §3.3].

### Bowden cable with sheath

The line runs inside a sheath, as in a Bowden cable. The review shows Bowden cable
actuation as one case of the drive for these robots [arxiv:1807.03731 §3.3,
fig. 6]. Here the friction comes from the sheath, and
it depends on how the sheath curves between the actuator and the end effector.
The review does not give a rule for choosing between the variants.

## Losses and how they are modelled

The review lists several friction models without preferring one [arxiv:1807.03731 §3.3].

- A simple transmission model assumes Coulomb friction. A dynamic model adds
  the viscoelastic properties of the tendon [arxiv:1807.03731 §3.3].
- For a Bowden pair, a capstan-type friction loss is approximated by constant
  radii of curvature along the sheath. Non-uniform curvature is handled with a
  continuous time-domain model of tendon-sheath transmission [arxiv:1807.03731
  §3.3].
- Conduit-induced friction has been added to the kinematic model of a
  multi-section tendon robot [arxiv:1807.03731 §3.3].

The review also reports ways to offset the losses:

- Recursive linear estimation, using measured actuator positions and the
  measured end-effector pose, can estimate backlash and correction factors for
  the elastic properties [arxiv:1807.03731 §3.3].
- A Fourier-series characterisation of the motion losses, combined with
  feed-forward compensation from a statics model, has been used on a
  multi-segment multi-backbone robot [arxiv:1807.03731 §3.3].
- A smooth hyperbolic-tangent inverse has been used as backlash feed-forward
  compensation in wire-actuated robots. For catheters, a dead-zone model of the
  backlash, combined with Coulomb friction, has been used [arxiv:1807.03731
  §3.3].

If the model is not enough, many works add an external measurement of the tip
to the control loop. Examples the review names are image tracking and
electromagnetic sensing [arxiv:1807.03731 §3.2].

## When to use which

The review does not rank these options and reports no measured comparison.
The table reads its section descriptions against the situation.

| Situation | Use | Why |
|---|---|---|
| The actuators must be kept away from a small moving part | Tendon or cable drive, with actuators placed remotely | This is the reason the review gives for remote actuation [arxiv:1807.03731 §3.3] |
| The tool needs jointed dexterity at its tip | Wire-actuated articulated end effector | The review cites this design for single-site and single-port systems [arxiv:1807.03731 §3.1] |
| The arm must bend along its length through a narrow path | Multi-backbone push-pull arm | This is the design the review gives for multi-segment bending [arxiv:1807.03731 §3.1, fig. 4] |
| The line runs through a sheath with bends | Model the sheath friction, such as with the Bowden or tendon-sheath models | The review lists these models for sheath curvature and friction [arxiv:1807.03731 §3.3] |
| Motor readings alone must give tip position | Compensate backlash and friction in the model | The review names friction, backlash and stretch as the causes of the error [arxiv:1807.03731 §3.3] |
| Tip accuracy must be better than the model allows | Add an external tip measurement to the loop | The review names image tracking and electromagnetic sensing for this [arxiv:1807.03731 §3.2] |

The review does not give numeric accuracy for any of these options, so the
table does not state how large the gains are.

## Implementation sketch

The review gives no code for these methods, and it does not state their
parameters. This sketch models backlash as a dead zone, which the review
reports for catheters [arxiv:1807.03731 §3.3]. It leaves out friction. The
dead-zone width is a free parameter that has to be measured for each line.

```pseudocode
# Backlash as a dead zone on the output of a tendon line
# Input:  commanded displacement u, previous output y_prev,
#         backlash width b (measured per line)
# Output: displacement y transmitted to the end effector
# Convention: the slack is split evenly, b/2 on each side (not from the review)

def transmit(u, y_prev, b):
    # The output moves only once the command leaves the slack band around
    # the previous output  [arxiv:1807.03731 §3.3]
    if u > y_prev + b / 2:
        return u - b / 2
    if u < y_prev - b / 2:
        return u + b / 2
    # Inside the slack: the output stays where it was
    return y_prev
```

## Contradictions

None known. The review lists several friction and backlash models for the same
kinds of line, and it does not compare them [arxiv:1807.03731 §3.3]. This is a
gap in its coverage, not a disagreement between sources.

## Papers

- [Medical Technologies and Challenges of Robot Assisted Minimally Invasive Intervention and Diagnostics](/papers/1807.03731): a review of confined-space surgical robots. It covers wire-actuated and continuum designs, the friction and backlash models for their lines, and the compensation methods reported for them.
