Instantaneous Center of Rotation (ICR) Analysis

• Instantaneous Center of Rotation (ICR) is a geometric construct that defines the point or axis about which a body rotates with zero instantaneous velocity, serving as a key element in nonholonomic kinematics.
• It underpins control and estimation techniques by guiding trajectory tracking, handling singularities, and integrating sensory data in both planar and 3D robotic systems.
• Practical applications of ICR include optimizing energy efficiency, enhancing motion estimation accuracy, and reducing system instability in both robotic mechanisms and biomechanical analyses.

The instantaneous center of rotation (ICR)—also referred to as the instantaneous axis of rotation or, in specific domains, the virtual pivot point (VPP)—is a geometric construct fundamental to the analysis of nonholonomic robotic mechanisms, biomechanical systems, and movement estimation. At each time instant, the ICR is the unique point (or axis in 3D) in a body or mechanism about which the system rotates with zero instantaneous velocity. Its identification underpins control, estimation, and analysis across robotics, human motion science, and kinematics.

1. Mathematical Formulations of the Instantaneous Center of Rotation

Planar mechanisms define the ICR as the unique point at which the velocity vector is momentarily zero in the plane of motion. For mobile robots with differential drive or complex steering mechanisms, the ICR for the platform is the intersection point of all the module-referenced steering axes. For the four-module steerable robot hTetro, individual module velocities $v_i$ and steering rates $\dot\beta_i$ relate to wheel angular velocities $\dot\phi_{iL},\ \dot\phi_{iR}$, and the forward kinematics map them to the body pose derivative via a steering-body Jacobian $G$: $\dot\xi^* = R(\theta^*)\,G(\alpha_j,\beta_i)\,\begin{bmatrix}v_1\v_2\v_3\v_4\end{bmatrix}$ where $\xi^*=[x^*,\,y^*,\,\theta^*]^T$ is the centroid pose, $R(\theta^*)$ is the world-to-body rotation, and $G$ encodes steering/drive geometry. Imposing that all modules’ wheel-normal lines intersect in a common $(x_{ICR},y_{ICR})$ determines admissible steering configurations (Shi et al., 2019).

In three dimensions, screw theory generalizes the ICR to the instantaneous axis of rotation. For a rigid-body twist $\xi=[\omega;v]$, with angular velocity $\omega$ and linear velocity $v$, the spatial instantaneous axis has direction $u=\omega/\|\omega\|$ and passes through the point $p=(\omega\times v)/\|\omega\|^2$ (Ziai, 2019).

In bipedal locomotion, the VPP (functionally equivalent to the ICR) is computed as the intersection of non-collinear ground-reaction force (GRF) lines: $r_i(\lambda_i) = p_i + \lambda_i d_i$ where $p_i$ is the center-of-pressure and $d_i$ the normalized GRF for each foot. The intersection $(x_{IP},y_{IP},z_{IP})$ satisfies $p_1+\lambda_1\,d_1 = p_2+\lambda_2\,d_2$ with closed-form for $\lambda_1,\lambda_2$ (Schreff et al., 2022).

2. The ICR in Control and Estimation of Robotic Systems

For self-reconfigurable robots such as hTetro, the ICR serves as the foundation for trajectory tracking and redundancy resolution. The central methodology uses a hierarchical controller: a high-level layer computes the desired ICR trajectory from global path errors; middle and low-level controllers distribute steering and drive commands to ensure all steerable modules’ axes remain concurrent or parallel as appropriate. Kinematic and representation singularities—when the ICR approaches infinity or aligns with a module’s axis—are handled by saturating the rotation radius and by failover to a special-case controller. The controller tunes the module steering rates and drive velocities such that the ICR constraint is enforced throughout transients, guaranteeing robust path following even across discontinuous waypoints and reconfiguration of the platform (Shi et al., 2019).

In skid-steering robots, the ICR model augments visual-inertial odometry by introducing a five-parameter random-walk model for the time-varying ICR $\xi=[X_v,Y_l,Y_r,\alpha_l,\alpha_r]^T$. The estimator fuses odometry, vision, and inertial measurements in a sliding-window optimization, reflecting the nonholonomic constraint that the robot’s velocity is instantaneously aligned with rotation about the ICR. The approach admits online adaptation to terrain-induced changes in skid/slip and ensures observability so long as nontrivial steering and yaw motions are present (Zuo et al., 2019).

3. The ICR in Biomechanical and Human Movement Analysis

In movement analysis, the ICR—often termed the center of rotation (COR) for joints—is crucial for biomechanics, exoskeleton control, and clinical kinematic assessment. Screw theory formalizes the ICR as the spatial axis about which instantaneous joint rotation occurs. The joint twist $\xi_i=S_i \dot\theta_i$, for screw axis $S_i$ and joint rate $\dot\theta_i$, sums along a kinematic chain to give total end-effector twist $\xi_{inst}$. The spatial ICR is then extracted as above from $\xi_{inst}=[\omega;v]$.

The IMU-based ArVE$_n$ algorithm provides real-time adaptive estimation of the joint COR using a single inertial sensor, embedding the IMU–COR vector and its time derivative into an extended Kalman filter (EKF). The measurement model exploits rigid-body kinematics: the IMU linear acceleration in its frame is

$$a_A = -\,\dot\omega\times r - \omega\times(\omega\times r)$$

where $\omega$ and $r$ are the estimated angular rate and IMU–COR vector. The system adapts to soft-tissue artifacts by tracking both $r$ and $\dot r$. In synthetic and in vivo experiments, ArVE$_n$ outperformed batch least-squares methods, achieving sub-10% error in joint-center location at 100 Hz rates (García-de-Villa et al., 2024).

4. Singularities and Robustness in ICR-Based Modeling

Two principal classes of singularities arise in ICR-based control:

• Representation singularity (ICR at infinity): Occurs in pure translation when the rotational component vanishes, i.e., $\dot\theta_d \rightarrow 0$ for hTetro or $\omega_z\to0$ for skid-steer robots. The rotation radius diverges, leading to ill-conditioning. Solutions include bounding $R_d$ with a saturation function, e.g., $R_d=R_{max}\tanh$ of the velocity/rotation ratio (Shi et al., 2019).
• Kinematic singularity (collinear/parallel axes): Occurs when two or more module axes become parallel, so the intersection is not unique. Mechanical design choices (decoupling steering and drive) and control algorithms (using PID plus scaling to enforce concurrency or parallelism) ensure the singular regime is detected and handled with an appropriate lower-rank solution (Shi et al., 2019).

Estimation approaches managing time-variation (random-walk prior on ICR parameters) and continuous residual enforcement further enhance robustness to parameter drift due to terrain, slippage, or soft-tissue artifacts (Zuo et al., 2019, García-de-Villa et al., 2024).

5. The Functional Role of the ICR/VPP in Human and Biological Locomotion

In human walking, the ICR concept generalizes to the virtual pivot point (VPP), representing the intersection point of GRFs. Ubiquity of VPP/focused ICRs (as measured by $R^2\to1$) has led to hypotheses regarding their role in stability. Schreff et al. (2023) demonstrate that postural stability in a neuromuscular human model can be achieved even when GRF intersection is absent ($R^2\lesssim0.6$)—both "IP gait" and "non-IP gait" models reject step-down perturbations equally well. However, absence of a coherent ICR increases the collision fraction (energy lost when the CoM velocity and GRF oppose), which raises mechanical cost of transport by $10$–$20\%$. Thus, the VPP/ICR in bipedal locomotion primarily improves energetic efficiency rather than being strictly needed for upright stability, and may additionally simplify supra-spinal control or reduce muscle/joint torque demands (Schreff et al., 2022).

6. Experimental Metrics and Performance Outcomes

Quantitative performance metrics for ICR-based control and estimation are directly reported:

• hTetro mobile robot: RMS path-tracking errors between $0.108\ \mathrm{m}$ (O-shape) and $0.129\ \mathrm{m}$ (S-shape) in $X$, $0.062$–$0.16\ \mathrm{m}$ in $Y$. Control system exhibits rapid recovery from waypoint jumps and disturbances, with minor overshoot attributed to subsystem tuning (Shi et al., 2019).
• Skid-steer robots: The ICR estimation rapidly adapts to terrain changes and slip, maintaining estimator validity across surfaces, with converged parameter estimates in seconds when excitation is present (Zuo et al., 2019).
• IMU-based joint COR estimation (ArVE$_n$): In synthetic "spherical pendulum" trials, ArVE$_n$ achieves $|\Delta r|=6.8\pm3.9\ \mathrm{mm}$ (error $<3.5\%$). In vivo hip-joint trials with 5 volunteers, average error is $9.5\%$ of the real IMU–joint vector value, with $4\pm1^\circ$ angular deviation; batch least-squares approaches exhibit roughly triple the error (García-de-Villa et al., 2024).

These results highlight the practical viability and limitations of ICR-based methods across different robotic and biomechanical systems.

7. Theoretical and Practical Limitations

ICR-based modeling requires that the critical kinematic assumptions are satisfied: for planar mechanisms, “no-slip, no-skid” is presupposed; for 3D rigid body applications, rigid link constraints and accurate state/parameter observability must hold. Soft-tissue artifacts, terrain-dependent slip, or structural compliance can degrade accuracy, although modern estimation frameworks (e.g., random-walk models for time-varying parameters) partially mitigate these influences. For human joint tracking, the approach presupposes a fixed or slowly moving COR, and may be inaccurate for joints with substantial translation.

Reliance on accurate orientation and velocity measurements is universal: misestimation of the IMU orientation, for instance due to magnetometer failure, can significantly impair filter accuracy (García-de-Villa et al., 2024). For estimation in visual-inertial odometry, degenerate motion patterns (e.g., constant speed/stationary or zero yaw rate) render ICR parameters unobservable (Zuo et al., 2019).

---

The ICR remains a central, unifying construct for nonholonomic control, kinematic modeling, and biomechanical analysis, with both theoretical depth and wide-ranging practical relevance. Its robust definition and estimation underpin stable and energy-efficient motion in robotics and biological systems alike.