Multi-Cable Actuation in Robotics

• Multi-Cable Actuation is a transmission architecture that routes multiple tendons through structures to generate precise forces and torques.
• It employs advanced mechanical designs, intricate cable routing, and nonlinear modeling to manage redundancy, tension allocation, and backlash minimization.
• Applications span continuum robots, dexterous hands, exosuits, cable-driven platforms, and shape-controllable space structures requiring versatile actuation control.

Multi-cable actuation is a class of transmission architectures in which multiple tendons or cables are routed through a structure to generate forces and torques for robotic actuation. This paradigm enables distributed, lightweight, and compliant actuation for continuum robots, dexterous hands, wearable exosuits, parallel manipulators, and shape-controllable structures. Multi-cable actuation encompasses both systems where each degree of freedom has a dedicated cable/actuator, and architectures where cables are differentially or multiplexedly mapped across multiple outputs, allowing tension allocation, redundancy, and efficient use of actuators. The complexity of routing, the nonlinearities of cable–structure interaction, and the variety of end-use scenarios motivate advanced mechanical designs, analytical and energy-based modeling, real-time tension optimization, and novel control architectures.

1. Mechanical Design and Cable Routing Strategies

Multi-cable systems employ intricate routing to achieve decoupled, redundant, or antagonistically coupled actuation. For soft manipulators, each modular section typically embeds evenly spaced, fiber-reinforced cable channels (e.g., at 120° for three-cable arms) within a soft elastomeric body, with rigid endcaps providing discrete anchor points and decoupling adjacent modules (Qi et al., 2024). In bionic hands, the routing often includes PTFE-lined sheaths, pulleys at critical joints (e.g., MCP and PIP), and extension springs to manage backlash and preload, spanning up to 30 tendons in a single hand (Han et al., 4 Dec 2025). For parallel robots, the entire workspace is enveloped by cables routed from fixed base winches to anchors on a moving platform, sometimes with looped cable arrangements to generate both translation and rotation (Métillon et al., 2021, Cheng et al., 2 Apr 2025).

Antagonistic pairings are common at joints, where flexor and extensor cables are routed in opposing directions through rolling-contact geometries, achieving force balance and backlash minimization even with remote-drive Bowden systems (Min et al., 31 Dec 2025). In continuum manipulators and aerospace deployables (e.g., solar sails), multi-cable actuation distributes control points along the structure—augmenting compliance and workspace coverage via spreaders, pulleys, or body-integrated guides (Qi et al., 2024, Lee et al., 23 Jan 2025).

Table 1: Representative Cable Routing Topologies

System Type	Number of Cables	Routing Complexity
Soft Arm (Qi et al., 2024)	3 cables/section	Embedded channels, rigid endcaps
Bionic Hand (Han et al., 4 Dec 2025)	30 tendons	Distributed pulleys, sheaths
CDPR (Métillon et al., 2021)	8 cables	Base-to-platform, looped routes
Humanoid Hand (Min et al., 31 Dec 2025)	30 Bowden	Remote drive, antagonistic
Solar Sail (Lee et al., 23 Jan 2025)	1–N per boom	Multiple spreaders, distributed

The physical layout directly influences friction, pre-tension requirements, compliance, routing-induced hysteresis, and achievable workspace or dexterity.

2. Mathematical Modeling and Static Equilibrium

Accurate prediction of how cable length changes or tensions map to system deformations is central to multi-cable actuation. Models must address geometric and material nonlinearities, routing-induced moment arms, cable-to-body penetration, and redundancy.

For soft manipulators without intermediate guides, a nonlinear static model is derived from equilibrium of moments and cable penetration into the soft matrix. For a section with n cables at radial offset d and angles β_i, the model is given by (see (Qi et al., 2024)):

$$\begin{cases} \sum_{i=1}^n T_i\,d\cos\theta_{0,i}\,\cos\beta_i = K_b\,\kappa_b \ \sum_{i=1}^n T_i\,d\cos\theta_{0,i}\,\sin\beta_i = 0 \ l_i = \frac{1}{\kappa_{c,i}} (L \kappa_b - 2\theta_{0,i}) \ T_i = \text{Defined by cable penetration and curvature} \end{cases}$$

Where $T_i$ are tensions, $\kappa_b$ is backbone curvature, and $\theta_{0,i}$ are cable incident angles. Numerical solution of this multi-variable nonlinear system yields the static configuration for given cable lengths.

Parallel or cable-driven continuum robots are often modeled by actuation-space energy methods, with total potential energy as:

$$E[q,a] = \int_0^L \frac{1}{2} (EI\kappa^2 + GJ\tau^2 + EA u^2)\, ds - \sum_{i=1}^n a_i \Delta\ell_i[q]$$

where $q(s)$ encodes local strain, and $\Delta\ell_i$ is the length change induced by the body deformation (Wu et al., 4 Sep 2025). Forward and inverse mappings leverage Hamilton’s principle and, in discretized form, are solved efficiently for complex cable routing.

Redundancy in parallel architectures is resolved by tension allocation within null-space constraints, often optimized for minimum norm or bounded tension (Bartels et al., 9 Mar 2026, Métillon et al., 2021). For differentially actuated systems, the steady-state mapping between input torque/angles and joint torques/positions is determined by the differential coupling matrix (Nobili et al., 14 Jun 2026).

3. Control Architectures and Tension Allocation

Control of multi-cable systems spans from direct position/torque control to advanced scheduling, tension optimization, and multiplexing.

In classical designs, each cable is regulated via closed-loop position or tension, with compensation for stretch, friction, and backlash (e.g., via tension-sensing or preloaded springs). For redundancy, desired joint torques are mapped to minimum-norm tension solutions using Jacobian pseudoinverses:

$T = (J^\top)^{+} \tau_{\text{des}}$

where $T$ is the vector of cable tensions, $J$ the Jacobian, and $T_i$0 the desired torque vector (Han et al., 4 Dec 2025).

Multiplexing and switching architectures (e.g., time-division multiplexing, mechanical switches, or electrostatic clutches) reduce actuator count by temporally or electrically disentangling control over multiple cables. For example, the MuxHand uses three BLDCs indexed over nine cables, with worm gears and electromagnetic clutches, updating each cable's tension at ~14 Hz while maintaining continuous-feeling actuation through high gear reduction and low-backdrivability (Xu et al., 2024). Similarly, electrostatic clutch-based multiplexers enable both time-division and simultaneous actuation of multiple cables from a single motor by engaging/disengaging clutches with sub-500 ms latency (Amish et al., 14 Jan 2025).

Haptic and force-rendering devices employ modular motor-brake modules, allocating desired endpoint forces to cable tensions via bounded optimization (e.g., Dykstra's algorithm) within hardware and passivity constraints (Bartels et al., 9 Mar 2026).

4. Trajectory Planning, Kinematics, and Inverse Solutions

Real-time motion planning in multi-cable actuated robots requires fast forward and inverse kinematic models that account for nonlinearities, redundancy, and constraints.

For multi-section continuum arms, the forward kinematics is constructed by concatenation of each section's homogeneous transform, governed by section curvature and orientation parameters. The Jacobian for task-space velocity, and its damped pseudoinverse, forms the core of numerical inverse kinematic solvers used in trajectory tracking (Qi et al., 2024).

In energy-based models, semi-analytic inverse schemes iteratively update actuation variables based on actuation Jacobian evaluations until spatial errors fall below tolerance (Wu et al., 4 Sep 2025). Optimization-based planners (as in TDMA) may use beam search to schedule actuation-space steps, optimizing travel, energy, and switching penalties under time-division constraints (Li et al., 18 Apr 2026).

Closed-loop feedback, including integral action on cable outputs or decentralized PID on joint encoders, is employed to reduce model–reality mismatch and stabilize high-DoF systems (Min et al., 31 Dec 2025, Gabellieri et al., 6 Mar 2025). In model-based differential architectures, sensorless torque estimation with friction compensation is performed in real time by inverting identified per-branch friction maps (Nobili et al., 14 Jun 2026).

5. Performance, Experimental Validation, and Trade-offs

Experimental characterization of multi-cable actuation systems spans workspace, force/torque capacity, precision, speed, and robustness. Key findings include:

• Soft modular arms (L=9.3 cm/section, d=1.25 cm) achieved φ_b up to 85°, workspace radius ~6 cm/section, tip tracking error reduction up to 52% with nonlinear static modeling versus baseline (Qi et al., 2024).
• Dexterous hands with distributed actuation (15 DoF, 1.4 kg, 30 tendons) demonstrated 11 N fingertip force, 10 kg power grasp, repeatability <0.5°, and stable performance over >1,000 cycles (Han et al., 4 Dec 2025).
• Bowden-cable antagonistic hands achieved 236 g distal mass, 18 N fingertip force, lifting >100× own mass with negligible trajectory deviation, exploiting optimized rolling-contact geometry to eliminate the need for motor synchronization (Min et al., 31 Dec 2025).
• Time-division and multiplexing architectures reduced actuator count by >50% while maintaining high payload-to-weight ratio (e.g., 10 kg payload with 2.17 kg manipulator, 1% end-effector accuracy under servo failures) and managing per-step actuation delays in sub-0.1s to sub-0.5s regime (Li et al., 18 Apr 2026, Xu et al., 2024, Amish et al., 14 Jan 2025).
• Modular cable haptic interfaces demonstrated up to 6 N active force, 186 N passive brake force, 20 Hz −3 dB bandwidth, and reconfigurability by module addition/removal (Bartels et al., 9 Mar 2026).

Design trade-offs arise among actuator count, mass/inertia, per-joint bandwidth, switching latency, redundancy, compliance, and backdrivability. Time-division and clutch-based multiplexing offer compelling means to scale DoF but introduce actuation delay and possible reduction in dynamic performance.

6. Applications and Emerging Directions

Multi-cable actuation is extensible across robot morphologies. Major application domains include:

• Soft continuum robots and manipulators, offering large workspaces and dexterity in unstructured environments (Qi et al., 2024, Wu et al., 4 Sep 2025).
• High-DoF biomimetic and humanoid hands for manipulation, grasping, and wearable/prosthetic systems (Han et al., 4 Dec 2025, Min et al., 31 Dec 2025, Xu et al., 2024).
• Cable-driven parallel robots and teleoperation masters, delivering high payload, large translational workspace, and haptic rendering (Métillon et al., 2021, Cheng et al., 2 Apr 2025, Bartels et al., 9 Mar 2026).
• Shape-control of flexible booms and space structures, notably solar sails (CABLESSail concept) (Lee et al., 23 Jan 2025).
• Exoskeletons and wearable robotics employing differential and switch-based architectures to minimize device weight and actuators while compensating for friction and providing redundancy (Nobili et al., 14 Jun 2026, Vadeyar et al., 7 Feb 2025).
• Aerial cable manipulation leveraging elastic-cable models and flatness-based trajectory planning for cooperative UAV/cable manipulation (Gabellieri et al., 6 Mar 2025).

Emerging research trends involve energy-based actuation modeling, optimization-based tension allocation, scalable and multiplexed drive schemes, robust and passivity-based feedback controls, and strategies to handle cable-induced non-idealities such as friction, compliance, and hysteresis. The modular, generalizable nature of multi-cable actuation enables adaptation to a wide spectrum of robot kinematics and task requirements.