Grasp Wrench Space

The grasp wrench space is formally defined as the convex hull of primitive wrenches. Each contact point between the gripper and object can generate a set of wrenches based on the friction cone model. The space is constructed in wrench space (ℝ⁶ for 3D objects), where axes represent three linear force components (Fx, Fy, Fz) and three torque components (τx, τy, τz). The Minkowski sum of wrenches from all contacts defines the total capability of the grasp.

A grasp achieves force closure if its wrench space contains the origin in its interior, meaning it can resist any external wrench. The quality of a grasp is quantified by its distance from losing closure. Common metrics include:

• Ferrari-Canny Metric: The radius of the largest sphere centered at the origin that fits inside the wrench space.
• Margin of Robustness: The minimum distance from the origin to the convex hull boundary.
• Volume Metric: The volume of the wrench space, indicating the general strength of the grasp. These metrics allow engineers to rank potential grasps for stability.

Real-world contacts are governed by Coulomb friction, creating a nonlinear friction cone of possible contact forces. To make wrench space analysis computationally tractable, this cone is often linearized into a polyhedral approximation. A common method is the pyramid approximation, which represents the cone with a set of m edge vectors. This transforms the continuous problem into a linear algebra one, where the wrench space becomes a convex polytope defined by the sum of these edge vectors from all contacts.

Wrench space analysis is the dual of grasp synthesis. Instead of searching for finger placements directly, planners evaluate candidate grasps by constructing their wrench spaces. Quality metrics computed from the space guide optimization algorithms (e.g., gradient descent, sampling) to find grasps that maximize robustness. This approach is central to analytic grasp planning and is used to generate training labels for data-driven methods like Dex-Net, where a neural network learns to predict grasp quality from point clouds.

While powerful, the classical wrench space model has assumptions that limit real-world application:

• Quasistatic Assumption: It typically ignores object dynamics and inertial forces.
• Known Geometry: Requires a precise 3D model of the object and its center of mass.
• Perfect Sensing: Assumes exact knowledge of contact points and friction coefficients.
• Rigid Bodies: Models the gripper and object as perfectly rigid, neglecting compliance and deformation. These limitations drive research into data-driven and hybrid models that complement the analytic approach.

Grasp wrench space analysis is specifically for force closure, which relies on friction. This contrasts with form closure, a stricter condition where the object is immobilized by the geometry of rigid contacts alone, without friction. A form-closed grasp will also be force-closed, but not vice-versa. In wrench space terms, form closure implies the primitive contact wrenches positively span the entire ℝ⁶ space, a stronger condition than just containing the origin. Most robotic hands rely on force closure due to the difficulty of achieving pure form closure with few fingers.

The Grasp Wrench Space is used to compute a grasp quality metric, quantifying the robustness of a grasp. The most common metric is the largest inscribed sphere within the wrench space, centered at the origin. A larger sphere indicates the grasp can resist a wider range of external disturbances. This metric is used to rank and select optimal grasp poses from a candidate set generated by vision systems.

• Example: A parallel-jaw gripper picking up a box. The GWS analysis might show a grasp on the box's center of mass yields a larger inscribed sphere than a grasp near the edge, making it the preferred choice for a stable lift.

A primary application is determining if a grasp achieves force closure. A grasp is in force closure if its Grasp Wrench Space contains the origin in its strict interior. This means the set of possible contact wrenches can generate any net wrench on the object, allowing the robot to counteract arbitrary external forces and torques. This is a binary, yes/no check for absolute grasp security.

• Key Insight: Form closure (immobilization via geometry) is a subset of force closure. A grasp can be in force closure even with fewer contacts than required for form closure, by using friction to apply forces within the friction cones.

Engineers use the Grasp Wrench Space to plan grasps that are specifically robust to expected task wrenches. Instead of just maximizing general robustness, the space is analyzed to ensure it contains the specific wrenches required for the manipulation task.

• Real Example: Screwing in a lightbulb. The task requires applying a significant torque about the axis of the bulb. A successful grasp plan would generate a GWS that extends far along that specific torque axis, even if the grasp is less robust to lateral forces. This is task-oriented grasping.

Constructing the exact, non-linear Grasp Wrench Space from friction cones is computationally complex. A standard practice is friction cone linearization, where the curved cone is approximated by a polyhedral cone (e.g., an 8-sided pyramid). The wrench space then becomes a convex polytope, allowing for efficient computation of quality metrics and force closure checks using linear algebra and computational geometry libraries.

The GWS is a critical evaluation module within modern grasp planning pipelines. A typical pipeline:

1. A perception system (e.g., a depth camera) generates a point cloud of a target object.
2. A sampler proposes thousands of candidate gripper poses.
3. For each candidate, a contact model estimates potential contact points.
4. The Grasp Wrench Space is constructed and evaluated for each candidate to compute a quality score.
5. The highest-scoring grasp is executed.

This process is central to frameworks like Dex-Net, which uses deep learning to predict grasp quality directly from visual data, implicitly learning a mapping to a robust wrench space.

The GWS framework is essential for designing and analyzing underactuated or soft robotic hands. These hands have fewer motors than degrees of freedom and use passive compliance to conform to object shapes.

• Application: For an underactuated hand, the set of achievable contact forces is constrained by the tendon routing and joint couplings. The GWS analysis reveals the specific subset of wrenches the hand can actually apply, which may be a lower-dimensional manifold within the full theoretical wrench space. This informs design trade-offs between simplicity, cost, and functional capability.

Force closure is a geometric and frictional condition where a set of contact forces can generate any resultant wrench (force and torque) on an object, ensuring it can be held securely against arbitrary external disturbances. It is the primary quality metric for a grasp.

• A grasp is in force closure if its Grasp Wrench Space contains the origin in its interior.
• This means the grasp can resist wrenches applied from any direction by appropriately modulating contact forces.
• Analysis involves checking the convex hull of primitive wrenches generated at each contact point.

Form closure is a stricter, frictionless condition where an object is immobilized purely by the geometry of rigid contacts, preventing any infinitesimal motion. It is a subset of force closure.

• Unlike force closure, form closure does not rely on friction; contact normals alone constrain all degrees of freedom.
• In the Grasp Wrench Space, a form-closure grasp can generate wrenches in all directions even with zero friction coefficients.
• Achieving form closure typically requires at least four frictionless contact points for a 3D object.

A friction cone is the set of all possible contact force vectors that can be applied at a point contact without causing slip, defined by the Coulomb friction model.

• The cone's axis is aligned with the surface normal, and its half-angle is arctan(μ), where μ is the coefficient of friction.
• Any force within this cone satisfies the non-slip condition. Forces on the boundary indicate impending slip.
• In wrench space analysis, the primitive wrenches for a contact are derived from the edges of its linearized friction cone pyramid.

In robotics and mechanics, a wrench is a six-dimensional vector that combines linear force and rotational torque acting on a rigid body. It is the fundamental quantity in grasp analysis.

• Represented as w = [f_x, f_y, f_z, τ_x, τ_y, τ_z]^T.
• The Grasp Wrench Space is the set of all net wrenches that can be applied on the object through the grasp contacts.
• External wrenches (e.g., gravity, inertial forces) must lie within the negative of the GWS for the grasp to resist them.

A contact model mathematically defines the interaction between a robotic manipulator and an object at a point of contact, dictating which forces and motions are possible.

• Point Contact with Friction (Hard Finger): Transmits both normal and tangential forces, bounded by a friction cone.
• Soft Finger Contact: Can also transmit a torque about the contact normal.
• Pure Rolling Contact: Allows no relative motion at the contact point.
• The chosen model directly determines the set of primitive wrenches used to construct the Grasp Wrench Space.

A grasp quality metric is a scalar measure that quantifies the robustness of a grasp configuration, often derived from the geometry of the Grasp Wrench Space.

• The most common metric is the minimum distance from the origin to the convex hull boundary of the GWS. A larger distance indicates a greater ability to resist disturbances.
• Other metrics include the volume of the GWS or its alignment with expected external wrench directions (e.g., gravity).
• These metrics are used to evaluate and rank candidate grasps generated by planning algorithms.