Zero-Moment Point (ZMP)

The Zero-Moment Point (ZMP) is formally defined as the point on the ground plane where the net moment of the inertial forces and gravitational forces acting on the robot has zero horizontal component. It is derived from the robot's full-body dynamics. The key equation is:

• Moment Balance: The sum of moments around the ZMP must be zero in the horizontal (x and y) directions: ( M_{x}(p) = 0, M_{y}(p) = 0 ).
• Calculation: It is computed from the robot's center of mass (CoM) acceleration, angular momentum, and the distribution of ground reaction forces (GRFs).

This makes the ZMP a dynamic measure, inherently dependent on the robot's motion, unlike static stability measures.

In practice, for a robot with multiple rigid foot contacts, the measured Center of Pressure (CoP) is physically identical to the ZMP. However, their conceptual origins differ:

• Center of Pressure (CoP): A kinematic point measured from the distribution of vertical ground reaction forces under the foot. It is where the resultant vertical force acts.
• Zero-Moment Point (ZMP): A dynamic point derived from the equations of motion.

Key Insight: When the robot is in static equilibrium or moving slowly, the CoP and ZMP coincide. During highly dynamic motions with significant angular momentum, the theoretical ZMP can exist outside the contact area, while the measurable CoP is always constrained within the support polygon. For stability, the ZMP/CoP must remain inside this polygon.

The primary stability rule derived from ZMP theory is that a legged robot is dynamically stable if the ZMP remains inside the convex hull of the contact points, known as the support polygon.

• Support Polygon: The smallest convex shape enclosing all points of contact between the robot's feet and the ground. For a single foot, it's the foot's outline.
• Stability Margin: The distance from the ZMP to the edge of the support polygon. A larger margin indicates greater robustness against disturbances.
• Implication for Control: Most ZMP-based walking pattern generators explicitly plan the robot's center of mass trajectory to keep the calculated ZMP within this polygon, often keeping it near the center for maximum stability.

While powerful, the ZMP criterion has specific limitations based on its underlying assumptions:

• Rigid Ground Contact: Assumes point contacts with non-deformable, high-friction ground. It does not model compliant or slipping contacts well.
• Simplified Dynamics: Classical ZMP-based planning often uses the Linear Inverted Pendulum Model (LIPM), which assumes a constant CoM height and ignores angular momentum.
• Conservative Gait: Enforcing a ZMP strictly inside the support polygon leads to statically stable, but often slow and inefficient, walking. It does not describe running or highly dynamic motions where the ZMP may legitimately leave the support polygon (e.g., during a flight phase).

These limitations motivate more advanced criteria like the Capture Point or Divergent Component of Motion (DCM) for faster, more agile locomotion.

ZMP is the cornerstone of the ZMP Preview Control method, the dominant algorithm for generating stable walking patterns for humanoid robots like Honda's ASIMO.

The standard pipeline is:

1. Footstep Planning: Determine where and when to place feet.
2. ZMP Trajectory Generation: Define a desired ZMP path that moves smoothly within the shifting support polygon.
3. CoM Trajectory Calculation: Using the LIPM dynamics, compute the CoM trajectory that will produce the desired ZMP path. This is often solved via Model Predictive Control (MPC).
4. Whole-Body Control: The desired CoM and foot trajectories are sent to a Whole-Body Controller (WBC) that computes joint torques to realize them.

This method provides a mathematically rigorous and verifiable approach to stability.

ZMP is one of several key metrics for legged stability, each with different applications:

• ZMP vs. Capture Point: The Capture Point is a point on the ground where the robot can step to come to a complete stop. It extends ZMP theory to account for the Divergent Component of Motion (DCM), enabling more reactive, push-recovery behaviors.
• ZMP vs. Angular Momentum: ZMP-based controllers often try to minimize centroidal angular momentum. More advanced Centroidal Dynamics planning explicitly manages angular momentum for more dynamic maneuvers (e.g., spinning).
• ZMP vs. Orbital Energy: For running models like the Spring-Loaded Inverted Pendulum (SLIP), stability is analyzed via orbital energy, a concept not captured by ZMP.

Engineering Choice: ZMP is preferred for planned, stable walking on known terrain, while Capture Point and centroidal methods are used for reactive locomotion and terrain adaptation.

The Linear Inverted Pendulum Model (LIPM) is the foundational simplified dynamic model used to derive the ZMP. It treats a walking robot as a point mass (the Center of Mass) atop a massless, telescoping leg. Key assumptions enable tractable analysis:

• Constant height of the Center of Mass.
• Motion confined to a horizontal plane.
• No angular momentum about the CoM. This linearization directly yields the ZMP equation: ZMP = CoM - (CoM_height / gravity) * CoM_acceleration. It is the primary model for generating dynamically consistent walking patterns in real-time controllers.

The Center of Pressure (CoP) is the point on a contact surface where the total Ground Reaction Force (GRF) vector is applied. It is a physically measurable quantity from force-torque sensors. For a flat foot in full contact with rigid ground, the ZMP and CoP are equivalent. However, they diverge in key scenarios:

• During foot roll or edge contact.
• On soft or deformable terrain.
• When significant rotational moments (e.g., from ankle motors) are applied. Thus, the ZMP is a theoretical stability point from a model, while the CoP is the actual force application point. A stable gait requires the ZMP/CoP to remain within the Support Polygon.

The Capture Point is a point on the ground where a robot can place its foot to bring its Center of Mass to a complete stop in one step. Derived from the LIPM, it accounts for the robot's current CoM velocity. The formula is: Capture Point = CoM + (CoM_velocity / sqrt(gravity / CoM_height)). It represents the Divergent Component of Motion (DCM), the unstable part of the dynamics. Modern reactive locomotion controllers use the Capture Point for:

• Push recovery: Calculating where to step to absorb a disturbance.
• Gait timing: Determining the latest possible foot placement. It extends ZMP-based planning by explicitly incorporating velocity for dynamic stopping.

Whole-Body Control (WBC) is a hierarchical optimization framework that coordinates all a robot's joints to execute multiple tasks simultaneously while respecting physical constraints. While ZMP-based planners generate a desired net force for the CoM, WBC distributes this demand across all limbs. It solves a Quadratic Program (QP) in real-time to compute joint torques that achieve:

• Primary tasks: ZMP/CoM trajectory tracking.
• Secondary tasks: Swing foot trajectory, posture control.
• Hard constraints: Joint limits, torque limits, friction cones. This allows a humanoid to, for example, maintain ZMP-based balance while reaching with an arm and avoiding self-collision, something a simple ZMP tracker cannot do.

Model Predictive Control (MPC) is an advanced control method that uses an internal dynamic model (like the LIPM) to predict future system behavior over a finite time horizon. At each control cycle, it solves an optimization problem to find a sequence of optimal control inputs (e.g., future ZMP locations or footstep positions) and applies the first step. For legged locomotion, MPC is superior to simple ZMP tracking because it:

• Anticipates future disturbances and plans preemptively.
• Explicitly handles constraints on footstep locations, ZMP, and actuator limits.
• Can optimize for smoothness and energy efficiency. It is the standard in modern humanoids for robust walking over uneven terrain.

A Reduced-Order Model (ROM) is a simplified mathematical representation that captures the essential dynamics of a complex system for planning and control. The ZMP is intrinsically linked to ROMs in legged locomotion. Key examples include:

• Linear Inverted Pendulum Model (LIPM): For walking (ZMP-based).
• Spring-Loaded Inverted Pendulum (SLIP): For running and hopping.
• Angular Momentum based models: For robots with significant torso or arm swing. The power of a ROM like the LIPM is that it reduces the high-dimensional dynamics of a humanoid (30+ joints) to a 2D point mass, enabling real-time ZMP calculation and gait generation. The control challenge is then to make the full robot faithfully execute the ROM's behavior.