Cable-Driven Parallel Robots (CDPR) & Tendon-Driven Manipulators — Robotics Reference

Cable-Driven Parallel Robots (CDPR) & Tendon-Driven Manipulators

1. At a glance

Cable-driven robots replace rigid links and gear trains with cables (or tendons) in tension as the load-bearing actuation element. A cable can only pull, never push; every cable-driven mechanism must therefore use redundancy, gravity preload, or antagonistic pairing to span the full task space. Two families dominate, and they share almost no design vocabulary except the cable itself:

• Cable-Driven Parallel Robots (CDPR) — N cables run from fixed base winches (motor + spool) to attachment points on a mobile end platform; varying cable lengths in a coordinated way moves the platform in up to 6 DOF. Workspace is defined by the convex hull of the anchors, not by the kinematics of a chain. Examples: SkyCam over an NFL stadium, Fraunhofer IPAnema, NIST RoboCrane, CableRobot Simulator (Max Planck), 3D-printing CDPRs for concrete.
• Tendon-driven serial manipulators — cables route along a serial kinematic chain via guides and pulleys, transmitting torque from base-located motors to distal joints. End-effector inertia is dramatically reduced; the entire actuator stack stays off the moving arm. Examples: Barrett WAM, Intuitive da Vinci EndoWrist, Shadow Dexterous Hand, Allegro Hand, Open Bionics Hero Arm, Festo BionicSoftHand.

Both classes win on the same axes: low moving mass (motors and gearboxes are stationary or remote), back-drivability (a cable + capstan is intrinsically compliant), long reach at modest cost, and the ability to put actuation outside a constrained or hostile region (inside the human body, inside a sterilizable shaft, behind a radiation shield, on the rim of a stadium). They lose on the same axes: unilateral force (cables can’t push — every controller must enforce $t_i>0$), cable wear and creep, friction nonlinearity (Stribeck + capstan), stretch-induced compliance (cables are springs at the bandwidths of interest), and routing complexity in serial chains. A CDPR additionally has cables that physically occupy the workspace and can collide with obstacles or each other.

Where it sits in the design stack: alternative to serial rigid-link arms and to parallel rigid-link machines (Stewart-Gough). The kinematic analysis borrows from parallel robots (Merlet 2006) but adds tension-distribution as a new design freedom. Backdrivability and tendon compliance push these systems into the admittance control regime by default — there is no fighting it.

First ask before choosing cable drive over rigid drive:

1. Is the workspace larger than any reasonable rigid arm can cover (stadium-scale, building-scale)? → CDPR.
2. Does the end-effector need to be very light (surgical instrument inside trocar, finger phalange, prosthetic)? → Tendon drive with remote motors.
3. Is back-drivability or inherent compliance essential (rehab, surgical force feedback, dexterous in-hand)? → Tendon drive.
4. Is the workspace 3D-printer-like (tall, big, fixed footprint, lightweight head)? → CDPR is dramatically cheaper than a gantry of the same envelope.
5. Can you tolerate non-pushing forces and design redundancy or gravity preload? If not, use rigid.

2. First principles

Cable tension constraint

A cable exerts only positive axial force: $t_i\ge 0$. Real cables additionally require a minimum pretension $t_{\min }$ (typically 5–50 N) to avoid slack and a maximum tension $t_{\max }$ set by yield or fatigue:

$t_{\min }\le t_i\le t_{\max },i=1,\dots ,N$

This single inequality reshapes every classical parallel-robot result: the wrench-feasible workspace (WFW) is the subset of the geometric workspace where, for every required end-effector wrench, there exists a tension vector inside the feasible box that produces it. WFW is always smaller than reachable workspace.

CDPR kinematics

Let $\mathbf{p}\in {\mathbb{R}}^3$ be platform position, $\mathbf{R}\in SO(3)$ orientation, ${\mathbf{a}}_i$ the $i$-th fixed base anchor (world frame), and ${\mathbf{b}}_i$ the $i$-th platform attachment (platform frame). The $i$-th cable vector and length:

${\mathbf{l}}_i={\mathbf{a}}_i-(\mathbf{p}+\mathbf{R}{\mathbf{b}}_i),L_i=\Vert {\mathbf{l}}_i\Vert ,{\hat{\mathbf{u}}}_i={\mathbf{l}}_i/L_i$

Inverse kinematics (given pose, find cable lengths) is therefore trivial — one Euclidean norm per cable. This is the opposite of serial arms.

Forward kinematics (given cable lengths, find pose) is a system of nonlinear equations; for a 6-DOF redundantly-constrained CDPR with N = 8 cables, it is over-determined and solved iteratively (Levenberg-Marquardt, Newton-Raphson) starting from a recent estimate. Worst-case computation: 100–500 µs on a 1 GHz controller.

The CDPR Jacobian (Verhoeven 2004) maps cable-length rates to platform twist; equivalently its transpose maps cable tensions to platform wrench:

${\mathbf{J}}^{\top }=\left[\begin{array}{ccc}{\hat{\mathbf{u}}}_1& \cdots & {\hat{\mathbf{u}}}_N\\ (\mathbf{R}{\mathbf{b}}_1)\times {\hat{\mathbf{u}}}_1& \cdots & (\mathbf{R}{\mathbf{b}}_N)\times {\hat{\mathbf{u}}}_N\end{array}\right]\in {\mathbb{R}}^{6\times N}$

The wrench balance is ${\mathbf{J}}^{\top }\mathbf{t}+{\mathbf{w}}_{\text{ext}}=0$, with $\mathbf{t}\in {\mathbb{R}}^N$ the cable tensions and ${\mathbf{w}}_{\text{ext}}$ the external wrench (gravity + load).

CDPR class by cable count

For a 6-DOF platform:

• Under-constrained ($N<6$): platform is gravity-stabilized; equilibrium pose depends on load. Used when gravity is reliable (cargo cranes, planar pick-place).
• Fully-constrained ($N=6$): platform is kinematically determined by lengths, but tensions are unique with no margin — any extra wrench requirement forces a cable through $t_{\min }$. Rare in practice.
• Redundantly-constrained ($N>6$, typically $N=7$ or $N=8$): cable-tension vector lies in an $(N-6)$-dimensional null space; the designer chooses a tension distribution inside the feasible box.

The Roberts–Graham–Trumpler condition (1998) states that at least 7 cables are necessary for full-rank wrench-closure of a 6-DOF platform without relying on gravity; 8 is the industry standard because it provides one cable of redundancy against single-cable failure.

Tension distribution

Given the wrench-balance constraint ${\mathbf{J}}^{\top }\mathbf{t}=-{\mathbf{w}}_{\text{ext}}$ with ${\mathbf{J}}^{\top }$ rank 6 and $\mathbf{t}\in {\mathbb{R}}^N$, the solution set is a 2-dimensional affine subspace for $N=8$. The standard pick is the minimum-2-norm tension inside the feasible box:

${\min }_{\mathbf{t}}\frac{1}{2}\Vert \mathbf{t}-{\mathbf{t}}_{\text{ref}}{\Vert }^2\text{s.t.}{\mathbf{J}}^{\top }\mathbf{t}=-{\mathbf{w}}_{\text{ext}},t_{\min }\le t_i\le t_{\max }$

solved as a small (8-variable) quadratic program at 1 kHz on a real-time controller (OSQP, qpOASES, quadprog). Gosselin & Grenier (2011) gave a closed-form interval method that avoids QP entirely for the $N=7$ case.

Tendon-driven serial chains — capstan equation

When a cable wraps around a stationary cylindrical guide (capstan), Euler’s belt-friction equation gives tension on the slack vs tight side:

$\frac{T_{\text{out}}}{T_{\text{in}}}=e^{-\mu \theta }$

with $\mu$ the friction coefficient (Teflon-coated steel on aluminium: 0.04–0.10; UHMWPE on polished steel: 0.06–0.12; Spectra in PTFE-lined sheath: 0.05–0.08) and $\theta$ the total wrap angle in radians, summed across all guides along the path. Long routes through several pulleys can lose 50–70 % of the source tension to friction — and the friction is hysteretic, so the loss reverses sign on the return stroke.

Antagonistic vs differential pairing

A single cable controls one joint in one direction. To get bidirectional torque on a revolute joint:

• Antagonistic pair — two cables pull in opposite directions on the same joint pulley. Net torque is $r(T_1-T_2)$; co-contraction $T_1+T_2$ sets joint stiffness independently. This is how human muscles work and how Shadow Hand and the Salisbury hand are tendoned.
• Differential — a single bidirectional cable loop drives the joint via a winch capable of paying out or reeling in. Simpler, half the cables, but no stiffness control.

The antagonistic configuration costs twice the cables and double the actuators but trades it for inherent variable impedance — a free impedance-control knob without admittance loops.

Sources of compliance

A real tendon is a spring-damper: stretch under load, internal friction during cycling. Stiffness:

$k_{\text{cable}}=\frac{E\cdot A}{L}$

For UHMWPE Dyneema SK78 ($E\approx 100$ GPa, much lower than steel’s 200 GPa), a 2 mm diameter ($A=3.14$ mm²) cable 1 m long has $k\approx 314$ kN/m — feels rigid at low load, becomes the dominant compliance once cumulative routing exceeds ~3 m. For a 7-DOF Barrett WAM with total cable path ~5 m per joint, the joint-level stiffness contribution from cable stretch is around $10^4$ N·m/rad — comparable to a harmonic drive.

3. Practical math + 3 worked examples

Example A — Planar 2-DOF CDPR point mass

Spec: 1 kg payload, two cables to ceiling anchors at ${\mathbf{a}}_1=(-2,4)$ m and ${\mathbf{a}}_2=(+2,4)$ m. Payload at origin. Cable working range $t\in [5,200]$ N.

Cable unit vectors from payload to anchors: ${\hat{\mathbf{u}}}_1=(-0.447,+0.894)$, ${\hat{\mathbf{u}}}_2=(+0.447,+0.894)$.

Wrench balance (planar, 2D — horizontal + vertical force only):

$T_1{\hat{\mathbf{u}}}_1+T_2{\hat{\mathbf{u}}}_2=\left(\begin{array}{c}0\\ mg\end{array}\right)=\left(\begin{array}{c}0\\ 9.81\end{array}\right)\text{ N}$

By symmetry, $T_1=T_2=T$; the vertical equation gives $2T\cdot 0.894=9.81$ N, so $T=5.49$ N. Just above the 5 N pretension floor — feasible but margin is small.

If the payload offsets to $(+0.5,0)$, the geometry skews: ${\hat{\mathbf{u}}}_1=(-2.5,4)/4.72=(-0.530,+0.847)$, ${\hat{\mathbf{u}}}_2=(1.5,4)/4.27=(+0.351,+0.937)$. Solving the 2×2 system:

$\left(\begin{array}{cc}-0.530& +0.351\\ +0.847& +0.937\end{array}\right)\left(\begin{array}{c}{T}_1\\ T_2\end{array}\right)=\left(\begin{array}{c}0\\ 9.81\end{array}\right)$

yields $T_1=4.16$ N, $T_2=6.28$ N. $T_1$ has dropped below the 5 N minimum — cable 1 has gone slack. Either raise the minimum-tension floor for both (over-tension solution) or constrain payload motion to a smaller WFW.

This 2D example shows in miniature what plagues every CDPR: the feasible workspace is a function of payload, orientation, and cable preload — not a fixed geometric volume.

Example B — 6-DOF CDPR, 8 cables, tension distribution

Spec: square room, 8 anchors on a 4 m × 4 m × 3 m ceiling grid (Cuboid pattern, 2 anchors per corner at different heights), 6-DOF mobile platform at the room centre $\mathbf{p}=(2,2,1.5)$ m. Platform mass 5 kg, payload 10 kg, no orientation offset. Cable working range $t\in [10,500]$ N.

External wrench (gravity only): ${\mathbf{w}}_{\text{ext}}=(0,0,-147.15,0,0,0{)}^{\top }$ N (and N·m).

Cable Jacobian ${\mathbf{J}}^{\top }\in {\mathbb{R}}^{6\times 8}$ is built by computing all 8 cable unit vectors and the moment arms $\mathbf{R}{\mathbf{b}}_i\times {\hat{\mathbf{u}}}_i$. With cuboid symmetry, ${\mathbf{J}}^{\top }$ has rank 6; the null space is 2D.

QP setup (minimum-tension distribution):

${\min }_{\mathbf{t}\in {\mathbb{R}}^8}\frac{1}{2}\Vert \mathbf{t}{\Vert }^2\text{s.t.}{\mathbf{J}}^{\top }\mathbf{t}=\left(\begin{array}{c}0\\ 0\\ 147.15\\ 0\\ 0\\ 0\end{array}\right),10\le t_i\le 500\text{ N}$

By symmetry of the centred pose, the unique minimum-2-norm solution distributes 18.4 N to each upward cable component; resolved across the 8 cables (4 carrying significant load, 4 with mostly lateral force), tensions cluster in the 12–25 N range — well inside the feasible box.

Solve time: OSQP at 1 kHz costs ~50 µs on an Intel NUC class controller. A typical industrial-grade CDPR controller (Fraunhofer IPA’s WireX on Beckhoff TwinCAT) hits 4 kHz with this QP solved every step.

Sensitivity: if the platform translates to a corner $(0.5,0.5,1.0)$, the asymmetry forces some cables toward 150–200 N and the corresponding antagonist cables toward $t_{\min }$. WFW boundary is hit when the QP becomes infeasible — i.e. no $\mathbf{t}$ exists in the box that balances the wrench.

Example C — Tendon-driven finger, capstan friction

Spec: 4-joint anthropomorphic finger (MCP, PIP, DIP, but four joints with MCP split), single tendon routed across 4 guide pulleys of radius $r=5$ mm each. Total wrap angle along the route is 320° = 5.585 rad. Friction coefficient $\mu =0.15$ for an uncoated Spectra cable in a polymer sheath.

Tension at the distal joint vs the proximal motor:

$\frac{T_{\text{distal}}}{T_{\text{motor}}}=e^{-\mu \theta }=e^{-0.15\cdot 5.585}=e^{-0.838}=0.433$

That is a 57 % friction loss from motor to distal joint on the pulling stroke. On the return stroke (driven by the antagonist tendon or by a return spring), the ratio reverses — the proximal motor sees only 43 % of the distal load it had to overcome.

Mitigation strategies with effect on the same 320° route:

Routing improvement	Effective μ	Loss
Uncoated Spectra in polymer sheath	0.15	57 %
PTFE-lined sheath	0.08	36 %
Ceramic-coated alumina guide pulleys	0.05	24 %
Ball-bearing supported pulleys (rolling)	0.01	5.4 %

Ball-bearing pulleys win every time but cost mass, volume, and complexity — fine inside a Barrett WAM forearm, infeasible inside a 7 mm da Vinci wrist. Surgical instruments thus accept high friction in exchange for compactness and compensate it with bilateral force-feedback estimation (Madhani 1998).

Required motor torque: to apply 5 N at the distal finger with the uncoated route, the motor must deliver $5/0.433=11.5$ N — and waste 6.5 N as heat in the cable and pulleys.

4. CDPR systems & design

Notable CDPR installations

• SkyCam (commercial since 1990, founded by Brown Engineering 1984) — 4-cable suspended-camera CDPR over US stadiums (NFL, MLB, NCAA football). Envelope up to 125 m × 100 m horizontal × 30 m vertical. Cables: Dyneema synthetic, 4–6 mm diameter; winches: servo-driven with constant tension control plus position synchronization. Direct competitor: Spidercam (Austrian-built, used by ICC cricket, FIFA, UEFA).
• NIST RoboCrane (Albus, Bostelman, Dagalakis 1989) — the foundational paper-and-prototype. 6-cable Stewart-platform crane; demonstrated suspended welding and large-part assembly. Pott (2018) cites this as the canonical reference design for fully-constrained CDPRs.
• IPAnema family (Fraunhofer IPA, Pott et al., 2010–present) — IPAnema 1/2/3 industrial CDPR for handling, 3D-printing, and assembly. 8 cables, 6-DOF, platform payload up to 200 kg, workspace up to 10 × 6 × 4 m. The reference platform for redundantly-constrained CDPR research.
• CableRobot Simulator (Max Planck Institute for Biological Cybernetics, Tübingen + Fraunhofer IPA, 2014) — 8-cable VR motion simulator, ~5 m diameter workspace, 350 kg cabin payload, ±0.5 g acceleration. Cabin houses a human passenger inside an HMD experiment.
• CoGiRo (LIRMM, Montpellier, 2013) — 8-cable redundantly-constrained CDPR, 15 m × 11 m × 6 m envelope, focused on large-workspace research and 3D-printing applications.
• FALCON-7 (Kawamura, Choe, Tanaka, Pandian, ICRA 1995) — 7-cable parallel wire-driven manipulator; published as one of the earliest high-speed cable-driven systems, demonstrated end-effector acceleration up to 43 g.
• WireMan (Williams, Ohio University, 2005) — 4-cable planar CDPR for material handling research.
• MARIONET-CRANE (LIRMM, 2009) — large-workspace rescue-and-recovery CDPR prototype.
• 3D-printing CDPRs — Bouygues + ETH Zürich for concrete extrusion (LCPP); Iridia for large-format additive manufacturing; CABLAR (Brückl 2019) for building-scale layered deposition.
• Rehabilitation CDPRs — NeReBot (Padua, Rosati et al.) for upper-limb post-stroke therapy; CableArm (Toronto); MACARM (multi-axis cartesian-based arm for upper-limb therapy, Mayhew et al.); SOPHIA (Switzerland).
• High-rise window-cleaning CDPRs — Tractebel and several Japanese / Korean systems; cables anchored at roof-edge winches, gondola constrained against window plane by additional tethering.

CDPR design parameters

Parameter	Typical range
Number of cables N	4 (under-constrained planar) to 8 (6-DOF redundantly constrained)
Cable material	Dyneema SK78/SK99, Spectra, Vectran, 7×19 steel rope
Cable diameter	1–8 mm depending on payload
Workspace volume	1 m³ (research) to 10⁴ m³ (stadium)
Platform payload	1 kg (research) to 1000 kg (concrete printing)
Cable speed	0.5–5 m/s sustained
Winch motor torque	5–500 N·m at the spool
Control update rate	1–4 kHz (real-time loop)
Position accuracy	±2–20 mm depending on cable creep
Repeatability	±1–10 mm

CDPR design heuristics

• Anchor pattern selection. Cuboid (cables to 8 corners of a box) is the default for 6-DOF redundantly-constrained CDPRs — it provides the largest centred WFW and is provably collision-free for cables in the central workspace. Pyramidal (4 anchors on a roof) suits suspended-camera under-constrained CDPRs.
• Platform attachment spread. Larger spread of ${\mathbf{b}}_i$ around the platform increases torque-arm leverage for moment generation — desirable for orientation control — but inflates collision volume. Industry rule: platform attachment spread ≥ 10 % of workspace dimension.
• Pretension floor selection. Choose $t_{\min }$ to keep all cables taut at the worst-case wrench and WFW corner. Too low → slack risk; too high → wasted cable load capacity and shortened cable life. Typical: $t_{\min }\approx 0.05\cdot t_{\max }$.
• Cable-to-platform mass ratio. When cable mass becomes > 5 % of platform mass, sagging of the catenary affects the kinematic model — use the “sagging cable” model (Irvine 1981) instead of the straight-line assumption.
• Winch placement vs anchor placement. For long cables, the winch should be co-located with the anchor — otherwise an additional fixed pulley introduces a friction-loss point and a routing-failure mode. Stadium CDPRs (SkyCam) co-locate the winch at the roof anchor.
• Cable life budget. Budget total cable cycles based on D/d, load, and material; replace at 70 % of L₁₀ life. For Dyneema SK78 at D/d = 12, $L_{10}\approx 2\cdot 10^5$ cycles at 30 % MBL (minimum break load).

5. Tendon-driven manipulators & grippers

Key tendon-driven robots

• Barrett WAM (Whole-Arm Manipulator) — Townsend & Salisbury, MIT/Barrett Technology, 1989–present. 4 or 7-DOF cable-driven arm. Motors live in the base; cables route through pulleys to drive distal joints. The signature design property is back-drivability without joint torque sensors — the motor can be felt directly through the cable. Used in 1000+ research papers; foundational platform for impedance control and human-robot physical interaction research.
• Intuitive Surgical da Vinci EndoWrist — 7-DOF wristed instrument that fits through an 8 mm trocar. All actuation via cables that route from a base disc (driven by the patient-side arm’s motors) through the instrument shaft to the wrist joints. Sterilized between cases. Patents US 6,594,552, US 7,806,891. The dominant cable-driven mechanism in volume — >7000 surgical robots deployed, each with multiple instruments, each with hundreds of cable cycles per case.
• Shadow Robot Dexterous Hand — Shadow Robot Company, UK, 1997–present. 24 movable joints (20 actuated DOF) driven by pneumatic muscles or by motor-pulled tendons; the human-hand reference for dexterous manipulation research.
• Allegro Hand (Wonik Robotics) — 16-DOF, 4 fingers × 4 joints, tendon-routed inside each finger. Lower cost than Shadow; popular for in-hand manipulation research at OpenAI, Berkeley, and TRI.
• Festo BionicSoftHand — pneumatic + tendon hybrid; demonstrates rolling and twisting in-hand manipulation; primarily a research demonstrator.
• Open Bionics Hero Arm — tendon-driven 6-DOF prosthetic, EMG-controlled, 3D-printed shell. First commercially successful tendon-driven prosthetic in volume.
• PSYONIC Ability Hand — cable + motor hybrid; first multi-articulating prosthetic with touch feedback to receive Medicare reimbursement (2023).
• Robotis OP3 hand and Robotis Manipulator-X — Dynamixel-driven, tendon for finger flex.
• Salisbury / Stanford-JPL Hand (Salisbury 1980) — the historical reference 3-finger 9-DOF tendon-driven gripper. Influenced every cable-driven hand since.
• MoonRanger / CMU steering — cable-actuated wheel steering on a lunar rover prototype (CMU 2022).

Tendon-driven design knobs

Knob	Range / Choice
Tendon material	UHMWPE (Dyneema SK78), Vectran, steel wire rope 7×19, polymer-coated tungsten (surgical)
Tendon diameter	0.3–1.0 mm (finger, surgical), 1.5–4 mm (arm)
Routing	Direct, single capstan, multi-pulley, sheathed (Bowden)
Pulley material	Aluminium (cheap), ceramic-coated alumina (low μ), PTFE bushings
Joint configuration	Antagonistic pair, differential, four-cable per joint
Pretension	5–50 N typical
Force sensing	Motor current, in-line load cell, capstan-arm strain gauge, FBG

Tendon-driven design heuristics

• Capstan wrap angle. A driven capstan needs ≥ 540° (1.5 turns) of wrap to prevent slip under transient load; below this, the static-friction coefficient is insufficient and slip occurs on impulse. Barrett WAM uses 720° (2 turns) plus a positive engagement set screw at the tail.
• Sheath vs guide routing. Open guides (ball-bearing pulleys) win on friction (μ ≈ 0.01) but lose on routing flexibility — guides constrain the cable to a fixed path. Bowden sheaths win on flexibility but pay 5–10× higher friction. Surgical instruments use a hybrid: guides at high-load zones (proximal capstan, joint pulleys), sheath in the flexible shaft.
• Antagonistic pair pretension trade. Pre-co-contraction $T_1+T_2=2T_0$ sets joint stiffness $k_{\theta }=2T_{0}r/r$ where $r$ is the pulley radius and $r$ the small-angle linearization (Hogan 1985). Higher pretension → stiffer joint and faster response, but proportionally higher motor torque burden and faster cable wear.
• Single-tendon-per-DOF with return spring. Half the cables (and half the actuators) at the cost of asymmetric dynamics — the spring fights every actuated motion. Acceptable for prosthetic fingers where natural-pose default is desirable; unacceptable for arms.
• Cable routing through joints. Routing a tendon across a joint axis without a coupling pulley introduces kinematic coupling — moving joint $i$ pays out cable to joint $j$. Avoid by aligning the cable with the joint axis (zero moment arm) or by using a dedicated idler pulley at the joint center.
• Take-up reservoir. A spring-loaded take-up reservoir at the motor end absorbs cable creep and accommodates thermal expansion without losing tension; mandatory for any tendon route > 1 m of synthetic rope.

6. Cable materials & components

Cable materials

Material	Tensile strength (GPa)	E (GPa)	Density (kg/m³)	Notes
Steel wire rope 7×19	1.6–2.2	200	7800	Heavy, fatigue-resistant, abrasive on guides
UHMWPE (Dyneema SK78)	3.5–4.0	100–120	970	Best strength-to-weight; creeps under sustained load; needs 10:1 D/d to avoid fatigue
UHMWPE (Dyneema SK99)	4.0–4.5	130–145	970	Premium variant, lower creep
Vectran (LCP)	3.0–3.2	75	1410	Heat-resistant; less creep than HMPE; used on rovers
Aramid (Kevlar 49)	3.6	124	1440	Heat-resistant but poor in compression and over small pulleys
Spectra (HMPE variant)	3.0–3.5	100	970	Comparable to Dyneema; common in marine + cable robots
Tungsten wire (surgical)	1.5–3.0	410	19300	Biocompatible, autoclavable, very small diameter feasible

Minimum capstan diameter (to avoid fatigue at $\ge 10^5$ bend cycles): D/d ≥ 10 for UHMWPE, D/d ≥ 25 for 7×19 steel, D/d ≥ 8 for Vectran.

Pulleys, guides, and routing

Component	Vendor / type	Application
Ball-bearing pulleys	Harken, Misumi, McMaster-Carr	Low-friction routing in tendon arms
Ceramic-coated pulleys (Al₂O₃, ZrO₂)	Customs (Misumi precision), Ondrives	Low μ in tight-routing contexts
Bowden sheaths (PTFE-lined)	Jagwire, Shimano (adapted), igus tribo	Long-route flexible tendon routing
Capstan drums	Custom-machined Al 7075 or steel	Winch in tendon arms
Winch drums + level wind	Custom (Fraunhofer IPA designs published)	CDPR cable storage
End fittings (swages, eye splices)	Petersen, Loos, Cousin Trestec	Anchor / platform attachment

Force sensing on cables

• In-line miniature load cell — Honeywell 31-series, Futek LSB200, ATI Mini-40 cable variant; 100 g to 500 N range; ~0.1 % FS accuracy.
• Motor current estimation — applicable when the gear ratio is small (winch directly on motor) and route friction is well-modelled. Cheap, no extra hardware. Used by SkyCam and most stadium CDPRs.
• Capstan-arm strain gauge — gauge bonded to a cantilever support of a guide pulley; measures the radial force on the pulley, related to cable tension by the wrap geometry.
• Fibre Bragg grating (FBG) — distributed strain sensing along the cable itself (Smartec, FiSens). Expensive but unrivalled for surgical and aerospace cables.
• Embedded conductive yarn — research-grade (UHMWPE braided with conductive yarn whose resistance varies with tension); useful for soft tendons in soft-robotics systems.

Winches and drums

A CDPR winch must coordinate cable speed (linear cable-out rate) with drum spool layering (level winding to prevent overlap). Standard architecture:

• Brushless servo motor (1–5 kW depending on payload) with planetary gearhead (3:1 to 30:1).
• Spool drum with helically grooved surface guiding the cable into a single layer at known radius.
• Level-wind mechanism (lead screw + cable guide) translating axially at exactly $\pi d_{\text{cable}}$ per drum revolution.
• Cable-tension load cell in-line between drum exit and routing guide.
• Renishaw absolute encoder on the drum (after gearbox) for direct cable-length feedback.

Drum radius affects cable-length resolution: a 100 mm drum with a 20-bit encoder gives sub-millimetre cable-length quantization, sufficient for ±2 mm platform repeatability in a 5 m CDPR.

7. Control challenges & failure modes

Control challenges

• Tension management — every command must satisfy $t_i\ge t_{\min }$. Slack means the cable can momentarily wrap, fold, or whip-crack on re-tensioning, which destroys cable life.
• Cable elasticity — cable is a spring with stretch up to 1–3 % at working load. The visible joint position lags the commanded position; the lag is load-dependent. Closed-loop position feedback at the platform (laser tracker, vision, IMU + Kalman) is therefore standard, not optional, for high-accuracy CDPRs.
• Cable inertia + propagation — for cables longer than ~10 m, wave propagation in the cable itself (speed ~1500 m/s in UHMWPE) becomes a noticeable lag at high-bandwidth control.
• Friction nonlinearity — capstan equation gives the equilibrium; real friction has Stribeck (velocity-dependent), pre-sliding hysteresis (Dahl, LuGre models), and temperature drift. Pavanelli & Pott (2017) cover identification.
• Cable hysteresis — slack take-up under direction reversal causes backlash-like response; quantifiable as $\sim 1$ mm dead zone per metre of cable for synthetic rope under nominal pretension.
• Workspace is task-dependent — WFW depends on payload, on direction of required acceleration, and on orientation. Planning has to account for this; off-line trajectory optimization is standard (CASPR has a WFW planner).
• Wrench-feasibility singularity — at certain orientations the cable-tension matrix ${\mathbf{J}}^{\top }$ loses full rank along a force direction; controller can’t deliver wrench, platform falls or sags. Pre-compute WFW boundary as part of commissioning.
• Cable-on-cable collision — particularly in spatial CDPRs as cables cross. Geometric check during planning; certain anchor patterns (Cuboid) provably avoid intersection in the central workspace.
• Cable-on-obstacle collision — the cables themselves are kinematic actors in the workspace; payloads with significant out-of-platform geometry need explicit clearance modelling.
• Cable break — single-fault failure that, in fully-constrained CDPRs, can drop the platform. Industry response: $N\ge 7$ redundancy with a tension distribution that keeps load tolerable on any 6 of the 8.
• Underconstrained CDPR perturbation — gravity-stabilized 3-cable CDPRs are dynamically stable only inside a region; perturbation beyond the stable manifold causes platform to swing into anchor or obstacle.

Failure modes

• Bend fatigue — repeated bend-over-pulley cycles cause fibre breaks. UHMWPE at D/d = 10 lasts ~10⁵ cycles; D/d = 20 lasts ~10⁶ cycles. Surgical instruments rotate cables through tight bends repeatedly; EndoWrist replacement schedule is “10 cases or 70 hours” per Intuitive’s published guidelines.
• Creep — synthetic ropes (HMPE family) creep under sustained load. SkyCam re-tensions weekly; rehabilitation CDPRs typically recalibrate cable length daily.
• UV degradation — outdoor synthetic ropes lose ~30 % strength in 12 months of direct UV. Coated or covered cables are mandatory for stadium installations.
• Sterilization damage — UHMWPE softens above 130 °C; standard autoclave at 134 °C destroys the rope. Surgical instruments use stainless or tungsten wire that survives autoclave; instrument shafts are designed for limited cycle life (~10 sterilizations) regardless.
• Stick-slip at low speed — high static-to-kinetic friction ratio causes jerk on motion start. Mitigation: small velocity bias, or pre-load that keeps cable always in motion at micro-scale.
• Sheath wear in Bowden tubes — repeated motion abrades the sheath; particles enter the cable and accelerate fatigue. Catheter robots replace the sheath as a serviceable item.
• Spool layering errors — if level-wind drifts, cable rides over itself, doubling effective drum radius and losing cable-length calibration. Mitigation: optical level-wind feedback (CDPRs published by Fraunhofer IPA include this).
• End-fitting failures — improperly swaged terminations are the dominant single-point failure mode in industrial CDPRs (Pott 2018). Hand-spliced eye-splices in UHMWPE retain 95–100 % of cable strength when done by certified rigger; mechanical swages on synthetic rope retain only 60–80 % and are not recommended.
• Temperature effect on E — synthetic rope modulus drops ~20 % between 0 °C and 40 °C; outdoor stadium CDPRs require thermal compensation in the cable-length model.
• Cable-tether weight — a 100 m cable of 4 mm UHMWPE weighs 1.2 kg; pre-tension to keep it from sagging into the workspace adds 50–100 N. Cable-tether weight cannot be ignored at stadium scale.
• Capstan slip — under sudden load (cable break elsewhere → wrench redistribution → spike on one cable), a friction capstan can slip; positive drive (sprocket, belt-and-pulley) avoids it but adds mass.

Reference data — typical numbers

Quantity	Typical range
Cable pretension floor $t_{\min }$	5–50 N (small CDPR), 50–500 N (stadium)
Cable working tension	0.1–0.3 × cable MBL
WFW volume / reachable workspace volume	30–70 % (depends on payload)
Catenary sag at 100 m, 500 N preload, 4 mm UHMWPE	~10 mm
Tension distribution QP solve time	20–100 µs (N=8, real-time controller)
Forward kinematics solve time (Newton-Raphson)	50–500 µs
CDPR position accuracy	±2–20 mm (cable creep + stretch)
Tendon arm position repeatability	±0.1–1.0 mm (cable stretch dominates)
Cable replacement schedule (commercial CDPR)	500–2000 working hours
Cable MBL safety factor	5:1 (industrial) to 10:1 (overhead-of-people)

8. Case studies

SkyCam — stadium-scale 4-cable CDPR

SkyCam, founded by Garrett Brown (inventor of the Steadicam) in 1984 and commercially operating since 1990, is a 4-cable suspended camera CDPR over a stadium. Cables are 4–6 mm Dyneema (specifically SK78 grade) routed from four corner winches at the upper rim of the stadium down to a camera reel weighing ~3.5 kg.

Engineering choices that have proven out over 35 years:

• Under-constrained 4-cable design (only 3 DOF actively controlled — X, Y, Z) — gravity preload is reliable for a downward-looking camera; orientation is handled by a gimbal on the platform, not by the cables.
• Constant-tension winch control — each winch runs a hybrid position-tension loop; under steady-state, the four tensions sum to balance gravity, but during fast moves the winches actively pull and pay out simultaneously.
• Synthetic cable over steel — the move from steel rope in early SkyCam to Dyneema in 1990s halved cable weight in the catenary and doubled cycle life.
• Cable-on-spectator-protection — modern SkyCam systems use force-limit cutouts that trip at <50 N additional load (snag detection).
• Maintenance schedule — full cable replacement every ~500 game-hours; tension recalibration daily; cable creep measurement after each game.

Workspace envelope reaches up to ~125 m × 100 m × 30 m at NFL stadiums. Acceleration is modest (~0.3 g) to keep the camera image stable for broadcast.

Intuitive Surgical da Vinci EndoWrist — tendon-driven 7-DOF in 8 mm

The EndoWrist instrument is a 7-DOF cable-driven manipulator with three wrist joints (yaw, pitch, roll) and a grasper degree of freedom, packed inside an 8 mm shaft and driven by cables that run from the patient-side arm’s input disc to the distal wrist.

Critical design decisions:

• All actuators are at the base — the patient-side arm has 4 motor-driven discs that interface with the instrument’s input pulleys. No motors in the sterile field, no electronics inside the patient.
• Cable material is tungsten wire — biocompatible, autoclave-survivable, and small enough (~0.3 mm) to fit four parallel routings inside an 8 mm shaft.
• Multi-cable per joint — each wrist joint is actuated by an antagonistic cable pair, with one redundant pair for the grasper; total 8 cables for the EndoWrist’s 4 disc inputs.
• Pulley routing — high friction is unavoidable (μ ≈ 0.2 around small steel pulleys); the surgeon’s force perception is therefore degraded, partially compensated by bilateral force estimation (Madhani 1998) and the surgeon’s visual feedback from the stereo endoscope.
• Cycle life — instruments are rated for ~10 uses; cable fatigue is the dominant retirement criterion. Intuitive uses an instrument-side encrypted memory chip to enforce the limit and bill per use.

By 2025, over 8000 da Vinci systems were deployed worldwide and ~14 million procedures had been performed; cable-driven instruments are the most-cycled cable robots in the world.

Barrett WAM — the cable-driven research arm

The Barrett WAM (Whole-Arm Manipulator) was developed by Bill Townsend at MIT in the late 1980s (Townsend & Salisbury, Cable-Driven Robots, 1989) and commercialized by Barrett Technology starting 1996. The 7-DOF version is the canonical cable-driven research arm.

Architecture:

• 7 brushless servo motors located in the base of the arm. None of the motors move with the arm.
• Each motor drives a cable-and-capstan transmission with a ~30:1 ratio. Cables are stainless steel wire, ~1.5 mm diameter, routed through ball-bearing supported pulleys.
• The motors couple to multiple joints through differential cabling — joint 1 and 2, joint 3 and 4, etc., are partially mechanically coupled via the cabling topology and decoupled in software.
• Backdrivability is approximately 70 % — friction is low enough that a human can push the arm by the end-effector and feel each joint’s motor as a back-EMF damper.
• Joint torque sensing is provided by motor current measurement — there are no strain-gauge torque sensors on the arm itself. This works because the cable transmission is reasonably efficient and the friction model is well-characterized.
• Payload: 3 kg at full reach; arm mass: ~27 kg; reach: ~1 m.

The WAM has been the basis of more than 1000 research papers on impedance control, haptics, surgical robotics, and human-robot interaction. Its influence on the rigid-arm cobot world — Franka Emika, KUKA iiwa, Kinova — is direct: those arms inherited the behaviour (backdrivable, torque-controllable) and reimplemented it with rigid links + harmonic drives + joint torque sensors.

Fraunhofer IPA IPAnema — industrial redundantly-constrained CDPR

IPAnema (Pott et al., Fraunhofer IPA, first generation 2010, current generation IPAnema 3 from 2017) is the reference industrial-grade CDPR — it sits between research prototypes (CoGiRo, CableRobot Simulator) and commercial niche systems (SkyCam, Spidercam) and has informed nearly every published industrial CDPR design over the last decade.

Architecture:

• 8 cables, 6-DOF redundantly-constrained. The cuboid anchor pattern is the published default and was extensively analyzed for cable interference and WFW shape in Pott (2014).
• Modular winch units. Each winch is a self-contained 1.5 kW brushless servo + 10:1 planetary + grooved-drum spool + level-wind + in-line load cell + Renishaw 22-bit absolute encoder on the drum shaft. Eight identical winch modules drive any 8-cable installation.
• EtherCAT control bus at 4 kHz update rate. The real-time controller (Beckhoff TwinCAT 3) runs the QP-based tension distribution + forward-kinematics estimator in the same cycle.
• Cable monitoring. In-line tension on every cable plus drum-shaft encoder feedback yields redundant length measurement; the difference detects creep, slip, or breakage in real time.
• Payload range 30–200 kg across IPAnema variants, with workspace envelopes from 4×3×2 m (IPAnema 1) up to 10×6×4 m (IPAnema 3).

The IPAnema design has been licensed and adapted for concrete-printing CDPRs (Bouygues), motion simulators (CableRobot Simulator), and warehouse pick-and-place research at Fraunhofer IPA’s industrial 4.0 lab. Pott’s WireX commercial software (2020) packages the IPAnema kinematic library for third-party CDPR designers.

Retrospective — why cable drive is still niche

Despite 35 years of commercial proof (SkyCam, da Vinci), 30 years of research-arm prominence (Barrett WAM), and 15 years of industrial CDPR development (Fraunhofer IPA), cable-driven robots remain a niche category compared to rigid serial arms. The reasons are honest engineering trade-offs:

• Force unilaterality doubles the cable + actuator count vs equivalent rigid arms.
• Cable creep and wear introduce a scheduled-maintenance burden that rigid harmonic / cycloidal transmissions don’t have.
• Tendon friction loss degrades the force-feedback fidelity that motivated tendon drive in the first place — there is a recursive design problem.
• End-fitting reliability is rigging-domain knowledge that robotics teams typically don’t have in-house.
• Cable physical occupation of workspace rules CDPR out of most workcell-replacement scenarios.

Conversely, the niches where cable drive dominates are exactly those where rigid arms can’t compete: stadium scale (SkyCam), trocar scale (da Vinci), high-DOF dexterous hand scale (Shadow / Allegro), large-build-volume 3D printing, and rehabilitation where compliance is a feature. The next-decade growth is most likely in hybrid architectures — rigid arms with cable-driven distal joints (humanoid hands on Optimus / Figure / Apollo), and CDPRs with rigid-arm secondary manipulators (Fraunhofer IPA’s published “CDPR + cobot” demonstrators).

9. Tools & software

Tool	Use
CASPR (UNSW, open-source)	MATLAB CDPR kinematics, workspace, control simulation
WireX (Fraunhofer IPA)	Industrial CDPR design + simulation (commercial)
CDPR-Sim / MotionSim	Simulation environments for stadium-scale CDPRs
CCDR-PROBA	Probabilistic workspace analysis tool (academic)
OSQP / qpOASES / MOSEK / IPOPT	QP solvers for tension distribution at control rate
MATLAB / Simulink	Dominant academic toolchain for CDPR + tendon research
ROS 2 + custom drivers	Production-grade controller stacks (Fraunhofer IPA, LIRMM publish ROS 2 nodes)
SolidWorks / Inventor	CAD for kinematic layout, anchor placement
Rhino + Grasshopper	Parametric design and visualization of cable routing (architectural CDPRs)
OpenSim / MuJoCo / Drake	Tendon-driven biomechanics + manipulation simulation

10. Edge cases & summary failure-mode table

Failure mode	Mechanism	Mitigation
Cable fatigue (bend-over-pulley)	Cyclic stress at wraps	D/d ≥ 10 for HMPE; D/d ≥ 25 for steel
Cable creep	Long-duration tensile load	Periodic re-tension; low-creep grades (SK99, Vectran)
Sheath wear (Bowden)	Friction abrasion	PTFE-lined sheath; serviceable replacement
UV degradation	Photodegradation of HMPE	Coated or covered cable; outdoor inspection
Sterilization damage	UHMWPE melt at 134 °C	Tungsten or steel wire; limited use cycle
Stick-slip at low speed	High μ_s/μ_k ratio	Low-friction routing; small velocity dither
Cable-to-cable abrasion	Crossing cables in tight routes	Pulley separators; routing planning
Spool layering error	Level-wind drift	Optical layering feedback
Spike on direction reversal	Slack take-up	Maintain pretension; differential drive
End-fitting slip	Improper swage on HMPE	Use spliced eyes; certified rigger
Capstan slip under shock	Insufficient wrap angle / preload	Positive drive (sprocket); 540° wrap minimum
Temperature elasticity drift	E(T) drops with temperature	Compensate in cable-length model
Cable break	Single-cable failure	$N\ge 7$ redundancy; load-shed control
WFW violation	Pose outside wrench-feasible region	Pre-computed WFW; trajectory replan
Capstan friction loss	Long pulley routes	Ball-bearing pulleys; ceramic guides

11. Cross-references

• manipulator-design — rigid-link arm design; the alternative architecture this note contrasts with.
• end-effectors — many cable-driven hands are end-effectors that bolt onto rigid arms.
• impedance-control — cable + capstan is intrinsically series-elastic; impedance / admittance control is the native control regime.
• sensors-force-tactile — in-line load cells, motor-current force estimation, FBG cable sensing.
• kinematics-dh — CDPR Jacobian construction shares math with rigid parallel manipulators.
• dynamics-rigid-body — platform inertia + payload dynamics enter the wrench balance.
• trajectory-generation — CDPR planners must respect WFW; rigid-arm planners assume static workspace.
• path-planning — cable-collision-aware planners for spatial CDPR.
• mobile-base-wheeled — alternative actuation for large workspace coverage.
• soft-robotics — tendon-driven soft robots use the same actuation principle with compliant body.
• mechanics-of-materials — cable as a uniaxial tensile element.
• materials-polymers — UHMWPE (Dyneema, Spectra) for synthetic rope.
• fasteners-bolts — end-fittings, swages, splice anchorage.
• fatigue-analysis — bend-over-pulley fatigue life.
• planned parallel-manipulators — Stewart-Gough and hexapod kinematics; CDPR borrows the parallel-robot Jacobian theory.

12. Citations

1. Pott, A. Cable-Driven Parallel Robots: Theory and Application. Springer Tracts in Advanced Robotics, vol. 120, 2018. ISBN 978-3-319-76137-4. The canonical CDPR textbook.
2. Bruckmann, T. & Pott, A. (eds.) Cable-Driven Parallel Robots — Proceedings of the First International Conference on Cable-Driven Parallel Robots. Mechanisms and Machine Science, vol. 12, Springer, 2013. ISBN 978-3-642-31987-7.
3. Gosselin, C. & Grenier, M. “On the determination of the force distribution in overconstrained cable-driven parallel mechanisms.” Meccanica, 46(1):3–15, 2011. DOI:10.1007/s11012-010-9369-x.
4. Verhoeven, R. Analysis of the Workspace of Tendon-based Stewart Platforms. PhD thesis, University of Duisburg-Essen, 2004.
5. Kawamura, S., Choe, W., Tanaka, S. & Pandian, S.R. “Development of an ultrahigh speed robot FALCON using wire drive system.” Proc. IEEE ICRA, vol. 1, pp. 215–220, 1995. DOI:10.1109/ROBOT.1995.525286.
6. Albus, J., Bostelman, R. & Dagalakis, N. “The NIST RoboCrane.” Journal of Robotic Systems, 10(5):709–724, 1993.
7. Townsend, W.T. & Salisbury, J.K. “The effect of transmission design on force-controlled manipulator performance.” MIT AI-TR-1054, 1988. (WAM precursor.)
8. Salisbury, J.K. “Active stiffness control of a manipulator in Cartesian coordinates.” Proc. IEEE CDC, pp. 95–100, 1980. (Stanford / JPL Hand cable design.)
9. Mason, M.T. & Salisbury, J.K. Robot Hands and the Mechanics of Manipulation. MIT Press, 1985. ISBN 978-0-262-13205-3.
10. Roberts, R.G., Graham, T. & Trumpler, P.R. “On the inverse kinematics, statics, and fault tolerance of cable-suspended robots.” J. Robotic Systems, 15(10):581–597, 1998.
11. Pavanelli, M. & Pott, A. “Empirical characterization of cable elongation in cable-driven parallel robots.” Cable-Driven Parallel Robots (3rd Int. Conf., Springer 2017), pp. 245–256.
12. Madhani, A.J., Niemeyer, G. & Salisbury, J.K. “The Black Falcon: a teleoperated surgical instrument for minimally invasive surgery.” Proc. IEEE/RSJ IROS, vol. 2, pp. 936–944, 1998. DOI:10.1109/IROS.1998.727320. (da Vinci EndoWrist precursor.)
13. Intuitive Surgical Inc. “Surgical tool with cable-driven articulating wrist.” US Patent 6,594,552 B1, 2003; “Cable-driven wristed surgical instrument.” US Patent 7,806,891 B2, 2010.
14. Rosati, G., Gallina, P. & Masiero, S. “Design, implementation and clinical tests of a wire-based robot for neurorehabilitation.” IEEE Trans. Neural Syst. Rehab. Eng., 15(4):560–569, 2007. (NeReBot.)
15. Merlet, J.-P. Parallel Robots, 2nd ed., Springer, 2006. ISBN 978-1-4020-4132-7. Parallel-manipulator theory shared with CDPR.
16. Williams, R.L. II “Planar cable-direct-driven robots, I: kinematics and statics.” Proc. ASME DETC, 2002. (WireMan precursor.)
17. Lamaury, J. & Gouttefarde, M. “Control of a large redundantly actuated cable-suspended parallel robot.” Proc. IEEE ICRA, pp. 4659–4664, 2013. (CoGiRo.)
18. Miermeister, P., Lächele, M., Boss, R., Masone, C., Schenk, C., Tesch, J., Kerger, M., Teufel, H., Pott, A. & Bülthoff, H.H. “The CableRobot Simulator: large-scale motion platform based on cable-driven parallel robot technology.” Proc. IEEE/RSJ IROS, pp. 3024–3029, 2016.
19. Dyneema Engineering Data — SK78 and SK99 rope specifications. DSM/Avient, 2023 datasheet.
20. Fraunhofer IPA — IPAnema product family. https://www.ipa.fraunhofer.de — IPAnema and WireX product brochures, current revision 2024.

---

Session log:

node ~/.claude/bin/obsidian-research.mjs log "Built Robotics/cable-driven-robots.md Tier 2 deep note"