Center-of-Pressure (CoP): Mechanics & Applications

• Center-of-Pressure (CoP) is the point on a support surface where the net reaction force is applied, critical for assessing load distribution and stability.
• It is quantified using methods like force plates, pressure insoles, and sensor arrays, with formulations that link contact mechanics to balance control.
• CoP is pivotal in biomechanics and robotics for applications such as gait analysis, humanoid control, and feedback systems in tactile manipulation.

Center-of-Pressure (CoP) is the point on a support surface where the resultant ground reaction force, or more generally the resultant contact force, can be considered to act. In biomechanics it is a standard quantity for postural control and gait analysis; in legged robotics it underlies balance criteria, walking pattern generation, and centroidal control; and in tactile manipulation it can be generalized to a compact descriptor of distributed contact. Across these domains, CoP links contact mechanics to stability because it encodes how load is distributed over the support region and how that distribution changes during stance, stepping, or object interaction (Funk et al., 2018, Ficht et al., 2023).

1. Mechanical definition and mathematical formulations

In biomechanics and balance control, CoP is the point of application of the ground reaction force vector on the support surface at which the net moment due to gravity and inertial forces is zero. In rigid-body terms, if the ground exerts a net wrench

$$\mathbf{w}= \begin{bmatrix} \mathbf{F}\ \mathbf{M}_O \end{bmatrix},$$

with $\mathbf{F}=(F_x,F_y,F_z)$ and $\mathbf{M}_O=(M_x,M_y,M_z)$, then for a flat-foot contact with normal aligned with the world $z$-axis and $F_z>0$, the planar CoP coordinates are

$$x_{\text{CoP}}=-\frac{M_y}{F_z}, \qquad y_{\text{CoP}}=\frac{M_x}{F_z}.$$

This formulation is explicit in humanoid-robot modeling and is also the force-plate interpretation of CoP in biomechanics (Ficht et al., 2023).

When pressure is sampled as a discrete distribution, CoP is the pressure-weighted average of sensor coordinates. If $P_i$ is the pressure at prexel $i$, located at $(x_i,y_i)$, then

$$x_{\text{CoP}}=\frac{\sum_i x_i P_i}{\sum_i P_i}, \qquad y_{\text{CoP}}=\frac{\sum_i y_i P_i}{\sum_i P_i}.$$

Equivalent formulations use local normal forces $\mathbf{F}=(F_x,F_y,F_z)$0 rather than pressures: $\mathbf{F}=(F_x,F_y,F_z)$1 These weighted-average definitions are used for pressure insoles, pressure plates, robotic foot sensors, and fingertip force arrays (Funk et al., 2018, Ruppert et al., 2020).

A generalized tactile formulation also appears in dexterous manipulation, where CoP denotes the resultant contact wrench reduced to a single force vector applied at a centroidal contact point. In that setting CoP comprises both a $\mathbf{F}=(F_x,F_y,F_z)$2 force vector and a $\mathbf{F}=(F_x,F_y,F_z)$3 contact point in a sensor frame; the authors explicitly describe this as a compact local contact descriptor rather than a complete representation of arbitrary multi-contact pressure distributions (Pan et al., 27 May 2026).

2. Relation to CoM, ZMP, CMP, and the support region

CoP is distinct from the Center of Mass (CoM). CoM is the location of the body’s mass distribution, projected to the ground when planar stability is considered; CoP is the point where the ground pushes back. Their relative positions are fundamental to postural control. In quiet standing, CoP trajectory reflects the control actions used to maintain balance, while in gait the CoP path reveals how load moves from heel to toe and from one foot to the other (Funk et al., 2018).

The support region is usually described as the Base of Support (BoS) or support polygon. In static balance, the CoM projection must lie inside the support polygon, and the CoP coincides with the CoM projection when no angular momentum is generated. In dynamic balance, CoP can move within the support polygon to shape CoM trajectory; stability criteria such as ZMP or CMP are then used. A recurring misconception is that CoM must always remain inside the support polygon during walking. The robotics formulations cited here state that, in dynamic walking, CoM can be outside the support polygon while CoP remains inside and the motion is stabilized via appropriate foot placement and centroidal control (Ficht et al., 2023).

In many robotics papers, CoP and Zero Moment Point (ZMP) coincide in planar coordinates under common assumptions, especially flat contact and negligible angular-momentum effects. When centroidal angular momentum is non-zero, the relevant pivot generalizes to the Centroidal Moment Pivot (CMP), with

$\mathbf{F}=(F_x,F_y,F_z)$4

or, equivalently, CMP as a CoP-offset quantity induced by upper-body torque. In that sense, ZMP/CoP is appropriate when angular momentum about the CoM is small or controlled to zero, whereas CMP extends the concept when upper-body dynamics are non-negligible (Nazemi et al., 2017).

Image-based stability work has operationalized these relations through two classic measures: $\mathbf{F}=(F_x,F_y,F_z)$5 and

$\mathbf{F}=(F_x,F_y,F_z)$6

with positive sign when CoM is inside the BoS and negative sign when it is outside. This suggests that CoP is most informative when interpreted jointly with CoM and BoS rather than as an isolated point (Scott et al., 2022).

3. Measurement and estimation modalities

Traditional CoP measurement uses force plates, pressure plates, pressure mats, or in-shoe plantar-pressure systems. Force plates derive CoP from measured ground reaction forces and moments; pressure insoles or plates compute CoP from the discrete pressure map. In low-cost posturography, a Wii Balance Board is treated as a force platform with four load cells and a board-centered CoP coordinate system (Kawasaki et al., 2024).

Recent work has extended CoP estimation well beyond direct force sensing. A rugged robotic sensor array, FootTile, estimates CoP on soft terrain from four miniature pneumatic force sensors; in a rolling-foot experiment its error in CoP $\mathbf{F}=(F_x,F_y,F_z)$7-coordinate was reported as $\mathbf{F}=(F_x,F_y,F_z)$8 mm during stance, compared with a commercial pressure plate (Ruppert et al., 2020). A wireless humanoid-foot system based on four load cells and an ESP32-C3 estimates CoP in real time and feeds a PID controller; experimental characterization reported an average measurement error of $\mathbf{F}=(F_x,F_y,F_z)$9 g, and the balance controller achieved a $\mathbf{M}_O=(M_x,M_y,M_z)$0 success rate in single-leg lifting tasks at a $\mathbf{M}_O=(M_x,M_y,M_z)$1-degree inclination with $\mathbf{M}_O=(M_x,M_y,M_z)$2 and $\mathbf{M}_O=(M_x,M_y,M_z)$3 (Muhtadin et al., 24 Dec 2025).

Indirect estimation from motion sensing is also feasible. An IMU-only statistical model for treadmill walking estimated pelvis-referenced global CoP from raw gyroscope, accelerometer, and magnetometer data, achieving average intra-subject RMS error of $\mathbf{M}_O=(M_x,M_y,M_z)$4 mm and average inter-subject RMS error of $\mathbf{M}_O=(M_x,M_y,M_z)$5 mm; the same study also showed that magnetometer channels could be dropped with only a small loss in performance and that the number of IMUs could be reduced to five without deterioration in model performance (Podobnik et al., 2020). Vision-based estimation has similarly reached millimeter-to-centimeter scales. In a Taiji dataset with synchronized video and insoles, PressNET inferred foot pressure maps from 2D pose and achieved mean CoP errors of $\mathbf{M}_O=(M_x,M_y,M_z)$6 mm for the left foot and $\mathbf{M}_O=(M_x,M_y,M_z)$7 mm for the right foot, substantially lower than a pose-space KNN baseline (Funk et al., 2018). Related image-based stability work reported mean CoP error of $\mathbf{M}_O=(M_x,M_y,M_z)$8 mm for fully image-based CoP localization and showed that pressure estimation remained the main bottleneck in fully image-based stability metrics (Scott et al., 2022).

Forecasting CoP before contact has also been demonstrated. On a level-ground to stair-ascent transition, a shank-mounted RGB-D system predicted the anterior-posterior foot CoP at impact with mean-absolute error of $\mathbf{M}_O=(M_x,M_y,M_z)$9 mm, $z$0 mm, and $z$1 mm at forecast horizons of $z$2, $z$3, and $z$4 ms, respectively (Murray et al., 9 Feb 2026).

4. CoP in balance control and locomotion

In humanoid locomotion, CoP often appears as the control input of the Linear Inverted Pendulum Model (LIPM): $z$5 where $z$6 is CoM position and $z$7 is CoP. This makes CoP the planar anchor through which ground reaction forces shape CoM motion (Ficht et al., 2023).

Reactive walking control commonly combines CoP modulation, step location and timing adjustment, and angular-momentum regulation. In a two-stage walking pattern generator, step timing and placement are first adapted from DCM measurements, then CoP and CMP are modulated to realize a desired DCM at the end of the current step while keeping CoP inside the support polygon (Nazemi et al., 2017). In position-controlled humanoids, ankle strategy is treated as CoP regulation and hip strategy as CMP regulation; the control objective is to compensate capture-point error by indirectly shifting CoP within the support polygon and, when necessary, moving CMP beyond it via upper-body angular momentum (Shafiee-Ashtiani et al., 2017).

CoP also supports model-free control. A proprioceptive humanoid framework plans pairs of CoP and foot-position objectives, searches around the current configuration by slightly moving leg joints while recording CoP and foot positions, and updates motion through optimization until all objectives are achieved. The approach requires no prior knowledge of kinematic or inertial parameters and exploits the quasi-static approximation in which ZMP and CoP are equivalent (Jiang et al., 2023).

At a finer level of analysis, static CoP sensitivity has been proposed as a further criterion for contact stability: roughly, the rate of change of CoP with respect to equilibrium configuration. This suggests that robustness is not captured solely by keeping CoP inside the support polygon; how rapidly CoP moves under small configuration changes can also distinguish balancing controllers (Romano et al., 2016).

5. Applications beyond laboratory gait analysis

CoP has become a target variable in human feedback systems. A haptic-guidance study used a Wii Balance Board to sense CoP and vibration motors around the abdomen to indicate the direction in which the participant’s CoP needed to be guided. Eight targets were placed $z$8 cm from the board center at $z$9-degree intervals, success required remaining within a $F_z>0$0 cm radius for $F_z>0$1 s, and the mean induction times were $F_z>0$2 s for haptic feedback, $F_z>0$3 s for visual feedback, and $F_z>0$4 s for auditory feedback; haptic and visual feedback showed no significant difference, while both were significantly faster than auditory feedback (Kawasaki et al., 2024).

An anticipatory gait-guidance system, ErgoTac-Belt, used vibrotactile cues to lead CoP during walking. The controller compared the current CoP with a future reference point $F_z>0$5 s ahead and activated front, back, left, or right tactors when the anticipatory error exceeded $F_z>0$6 m. Experiments on ten healthy subjects showed that vibrotactile guidance could keep CoP close to a predefined reference path with similar performance to visual feedback, including with eyes closed (Lorenzini et al., 2022).

In robotic manipulation, CoP is increasingly used as a local contact-stability variable rather than only a foot-ground quantity. Compliant gripping pads for a NAO humanoid measured normal force and CoP with three load cells; a calibration model reduced CoP mean absolute error from about $F_z>0$7 mm to about $F_z>0$8 mm, and the calibrated CoP was used to align pad surfaces and parameterize limit-surface-based gripping-force selection (Han et al., 2024). A multifingered force-aware controller for humanoids defined CoP on a virtual support plane as the weighted average of fingertip projections, using tactile-estimated normal forces as weights, and drove it toward the centroid of the fingertips contact polygon; the reported success rate was $F_z>0$9 on a balancing task and $$x_{\text{CoP}}=-\frac{M_y}{F_z}, \qquad y_{\text{CoP}}=\frac{M_x}{F_z}.$$0 in multi-object scenarios (Marra et al., 9 Mar 2026). In tactile sim-to-real dexterous manipulation, CoP has even been promoted as a physics-grounded contact representation, with policies conditioned on CoP achieving zero-shot sim-to-real transfer on peg-in-hole insertion and ball balancing (Pan et al., 27 May 2026).

6. Limitations, misconceptions, and open directions

Several recurring limitations are explicit in the cited work. Pose-based CoP inference has been demonstrated mainly on slow, controlled Taiji with six subjects; the authors note activity and population specificity, reliance on $$x_{\text{CoP}}=-\frac{M_y}{F_z}, \qquad y_{\text{CoP}}=\frac{M_x}{F_z}.$$1 pose, single-camera viewpoint, lack of explicit dynamics, fixed footwear and surface, and inter-subject variation as limitations (Funk et al., 2018). Fully image-based stability estimation remains constrained primarily by pressure prediction quality rather than CoM or foot-localization accuracy (Scott et al., 2022). Visual forecasting of foot CoP has so far been demonstrated only for anterior-posterior CoP on a level-ground to stair-ascent transition, with subject-specific models and fixed stair geometry (Murray et al., 9 Feb 2026).

In robotics, model-based CoP reasoning without force sensors depends on rigid, flat, no-slip contact assumptions and on correct support-phase identification (Ficht et al., 2023). A related conceptual caveat appears in tactile manipulation: the generalized CoP representation is explicitly a compact approximation, not a complete representation of arbitrary multi-contact pressure distributions (Pan et al., 27 May 2026).

Two misconceptions are especially persistent. First, CoP is not CoM. CoM is the body’s mass center; CoP is the effective application point of the support reaction. Second, CoP and ZMP are not universally interchangeable: they coincide under common flat-contact assumptions, but CMP is needed when non-zero centroidal angular momentum materially changes the effective pivot (Nazemi et al., 2017).

The reported future directions are correspondingly consistent. They include incorporation of $$x_{\text{CoP}}=-\frac{M_y}{F_z}, \qquad y_{\text{CoP}}=\frac{M_x}{F_z}.$$2 pose, temporal models such as RNNs or $$x_{\text{CoP}}=-\frac{M_y}{F_z}, \qquad y_{\text{CoP}}=\frac{M_x}{F_z}.$$3 CNNs, explicit modeling of additional dynamic quantities beyond CoP, subject-independent forecasting models, and real-time deployment on commodity hardware (Funk et al., 2018, Murray et al., 9 Feb 2026). A plausible implication is that CoP will continue to function as a unifying contact variable across biomechanics, humanoid balance, wearable feedback, and tactile manipulation precisely because it compresses distributed contact into a form that remains mechanically interpretable.