3D Wrist Rotations in Biomechanics & Robotics

• 3D Wrist Rotations are defined as coupled, multi-axial angular motions in anatomical and robotic wrists, integrating flexion/extension, radial/ulnar deviation, and forearm rotation.
• They are measured and modeled using advanced imaging, kinematic chains, and rotation representations like Euler angles and quaternions to ensure precision in motion analysis.
• Applications include human–robot interaction, assistive devices, prosthetic control, and rehabilitation systems, advancing both technological innovation and clinical practice.

Three-dimensional (3D) wrist rotations comprise the coupled, multi-axial angular motions enabled by anatomical, robotic, or biomechanical wrists and their synthetic analogues. In human morphology, the wrist complex allows for flexion/extension (FE), radial/ulnar deviation (RUD), and forearm pronation/supination (PS). Accurate sensing, modeling, actuation, and control of 3D wrist rotations are central to diverse research areas, including human–robot interaction, assistive and prosthetic devices, robotics, rehabilitation systems, and 3D perception pipelines. Leading approaches span anatomical analysis via imaging, parallel and serial robotic wrist architectures, origami-inspired exoskeletons, and machine learning estimators for 3D joint orientation.

1. Anatomical and Imaging Foundations of 3D Wrist Rotations

The anatomical wrist comprises a compound joint with axes of motion that do not remain fixed, instead exhibiting coupled 3D kinematics due to the motion of carpal bones and ligamentous constraints. High-resolution four-dimensional (4D) imaging—such as dynamic, fat-suppressed 3D MRI volumes—enables in vivo quantification of these rotations (Zarenia et al., 2021). Carpal bones (e.g., scaphoid, capitate, lunate) are segmented in both static and dynamic slabs, after which 6-DOF rigid registration (via iterative closest point alignment of bone surface point clouds) yields frame-wise $3 \times 3$ rotation matrices. Euler angles, specifically Z–Y′–X″ sequences, are extracted as

$$\begin{aligned} \phi &= \text{atan2}(r_{32}, r_{33}) \ \theta &= \arcsin(-r_{31}) \ \psi &= \text{atan2}(r_{21}, r_{11}) \end{aligned}$$

relating to FE ($\phi$), RUD ($\theta$), and PS ($\psi$). Peak ranges measured are $\sim 28.6^\circ$ (ulnar-radial) and $\sim 27.0^\circ$ (flexion-extension), with registration errors $\lesssim 4^\circ$. This quantifies the intrinsic nonorthogonality and translation of wrist axes, delineates physiological motion limits, and informs the design of robotic wrists and exoskeletons (Zarenia et al., 2021).

2. Kinematic and Dynamic Models for Robotic 3D Wrists

Robotic wrists achieve 3D rotations via serial, parallel, or hybrid kinematic chains, often with mechanical strategies inspired by anatomical findings.

Parallel and Hybrid Architectures:

Recent variable-stiffness (VS) designs employ 2-DoF parallel mechanisms for FE and RUD combined with a serial PS actuator to realize anthropomorphic 3D orientation. The minimal coordinate system $u = [u_1, u_2]^\top = [$RUD, FE$]^\top$ suffices for PM orientation; PS ($\varphi$) is controlled independently (Milazzo et al., 4 Dec 2025, Milazzo et al., 2023). Kinematics are derived by solving the closure equations of three-leg 4-R chains for the coupler and intersecting chain constraints, with closed-form inverse kinematics (IK) yielding all joint configurations for a desired pose. The general coupler orientation (using Z–Y–X Euler angles) is

$$T_{b \to c}(x) = T_z(x_c)\, T_y(y_c)\, T_x(z_c)\, R_z(\alpha_z)\, R_y(\alpha_y)\, R_x(\alpha_x)$$

Forward and differential kinematics, Jacobians, and joint-space dynamics (Euler–Lagrange) are explicitly computed for real-time control.

Serial and Spherical Wrist Actuators:

ByteWrist and DexWrist exemplify compact, torque-transparent parallel wrists that deliver independent 3-DoF roll–pitch–yaw (RPY) using nested three-stage arc linkages and ball-supported pivots (Peticco et al., 1 Jul 2025, Tian et al., 22 Sep 2025). The workspace is described in RPY angle space, and singularities are analytically characterized—parallel singularities arise when all linkage axes coplanarize; gimbal lock is circumvented by workspace limits.

Dynamic Axes and Human–Robot Alignment:

To capture non-fixed anatomical wrist axes, improved models introduce a floating/prismatic joint $d_2$ whose value is regressed as a nonlinear function of FE and RUD angles to track the moving carpal rotation center (Yu et al., 2020). The full rigid transform chain from proximal radius to fingertip allows real-time computation of anatomical axes, reducing misalignment and kinematic errors in human–robot interaction contexts.

3. Sensing, Representation, and Estimation of 3D Wrist Rotations

Measurement and representation of 3D wrist rotations for tracking and control employ various frameworks:

Rotation Representations:

Common parameterizations include Z–Y–X or RPY Euler angles, $3 \times 3$ SO(3) rotation matrices, unit quaternions, and reduced representations such as axis–angle or the Gibbs–Rodrigues vector $\mathbf{g} = \mathbf{u} \tan(\theta/2)$ [0104016, (Ludwig et al., 14 Apr 2025)]. Gibbs–Rodrigues offers minimal parameter count, analytic conversion to/from matrices, and efficient composition, but exhibits singularities at $|\theta| = \pi$.

Representation	Parameters	Noted Advantages
Euler angles	3	Intuitive, human-like axes
Quaternion	4 ($\|q\|=1$)	No gimbal lock, efficient interpolation
Axis–angle	3	Compact, direct axis interpretation
Gibbs–Rodrigues	3	No normalization, simple algebraic composition

Sensor-based Estimation:

Optical systems (e.g., Leap Motion, motion capture) provide full 6D hand/fingertip tracking, which—combined with anatomical or robotic kinematic models—allows inverse computation of all rotational DoFs (Yu et al., 2020).

Neural Estimation:

Approaches such as Hand4Whole (Moon et al., 2020) and transformer-based pose uplifting (Ludwig et al., 14 Apr 2025) directly estimate wrist rotations as axis–angle, matrices, or quaternions from 2D/3D joint observations and local features. Incorporation of metacarpophalangeal (MCP) joint cues yields a mean per-vertex positional error (MPVPE) improvement of $>$10 mm in hand orientation estimation. Mean per-joint angular errors for wrists $<9^\circ$ have been achieved on whole-body mesh benchmarks, at computational speeds $>150\times$ faster than inverse-kinematics baselines.

4. Actuation, Stiffness Regulation, and Compliance

Variable-stiffness implementations modulate end-point compliance through nonlinear elastic elements in the wrist drive train (Milazzo et al., 4 Dec 2025, Milazzo et al., 2023). Redundant elastic actuators permit real-time adjustment of global Cartesian compliance ellipsoids via internal torque control in the nullspace of the actuation-to-wrench Jacobian ($T_o = N\beta$; $N$ spans the nullspace, $\beta$ is a scalar stiffness command). To meet a target compliance $C^*$, scalar $\beta$ is updated online (Levenberg–Marquardt optimization). Disturbance rejection is tunable: a 1.5 kg load causes $>20\times$ less angular deviation and $>20\times$ higher restoring torque in rigid ($\beta$ high) vs compliant ($\beta$ low) modes, mimicking human cocontraction and enabling both precision and safe interaction.

5. Origami and Soft-Wearable 3D Wrist Rotation Devices

Kresling origami-inspired orthoses realize programmable, omnidirectional 3D wrist rotations through compliant, fabric-based cellular architectures with tendon-driven actuation (Liu et al., 30 Jan 2025). Module geometry is individualized by mapping anatomical circumference and length to origami cell parameters ($a_1,a_2,b,\alpha$), guaranteeing semi-folded neutral states and omnidirectional folding. Kinematic mappings link tendon shortening $\Delta\ell$ to joint angles ($\beta_{dor/pal},\beta_{rad/uln}$), enabling FE, RUD, compound dart-throwing, and circumduction trajectories of $18.8^\circ$–$32.6^\circ$ per tendon and up to $\sim 31.7^\circ$ for combined pulls.

6. Control Strategies and Computational Considerations

Closed-loop wrist rotation control integrates kinematic inversion, dynamic compensation, and, in advanced contexts, direct stiffness regulation. Example implementations:

• Dual-loop control in VS wrists: Position (orientation) is regulated via kinematic inversion and low-level PD/P control, while net end-point stiffness is controlled by adjusting internal spring preloads to track desired compliance tensors, decoupling orientation and stiffness regulation (Milazzo et al., 4 Dec 2025).
• Human-like path-independent movement: Spherical projection-based passive motion paradigms (PMPs) rely on impedance attractors (fractal impedance controller) on the orientation manifold to generate globally stable, low-complexity control without iterative optimization. This scheme affords real-time operation at $1$ kHz regardless of system redundancy (Tiseo et al., 2021).
• Jacobian-based velocity mapping: ByteWrist and similar robotics devices numerically estimate $3\times3$ Jacobians ($$J = \partial(\alpha,\beta,\gamma)/\partial(\theta_1,\theta_2,\theta_3)$$) for rate control and singularity detection. Parallel singularity avoidance is ensured by workspace and Jacobian conditioning constraints (Tian et al., 22 Sep 2025).
• Robustness and failure modes: Quaternion- and axis–angle–based learning methods for wrist rotation can diverge in the presence of ambiguities (e.g., quaternion double cover, $2\pi$ wrapping), especially in rapid or extreme supination/pronation. Joint feature fusion, keypoint selection, and VPoser priors enhance physical plausibility and mitigate implausible mesh configurations (Ludwig et al., 14 Apr 2025).

7. Comparative Perspectives and Application Domains

Robotic wrists with decoupled, human-like axes (FE, RUD, PS) and low-inertia, backdrivable designs (e.g., DexWrist) significantly expand reachable workspaces, increase maneuverability in constrained environments by $>2\times$ compared to serial wrists, and improve data efficiency in learning-based manipulation (Peticco et al., 1 Jul 2025, Tian et al., 22 Sep 2025). Human–robot alignment, facilitated by dynamic-axes models and precise compliance tuning, supports dexterous physical interaction, teleoperation, and prosthetic feedback (Yu et al., 2020, Milazzo et al., 2023).

In orthotic and rehabilitation applications, personalized, soft origami-based wrists enable tunable 3D motion patterns, fine workspace adaptation, and the reproducible synthesis of naturalistic compound trajectories (Liu et al., 30 Jan 2025). In 3D vision and biomechanics, accurate wrist rotation estimation—whether from imaging, direct sensing, or learning—remains foundational to whole-body pose recovery, motor function analysis, and the validation of robotic and orthotic designs (Moon et al., 2020, Ludwig et al., 14 Apr 2025, Zarenia et al., 2021).

References:

(Milazzo et al., 4 Dec 2025, Liu et al., 30 Jan 2025, Moon et al., 2020, Tiseo et al., 2021, Ludwig et al., 14 Apr 2025, Zarenia et al., 2021), [0104016], (Yu et al., 2020, Milazzo et al., 2023, Peticco et al., 1 Jul 2025, Tian et al., 22 Sep 2025)