Form closure: trapping an object with geometry
When a robot picks something up, the deep question is simple: can the object get away? A grasp that looks tight can still let a part slip, twist, or pop out the moment the arm accelerates. Engineers answer this with two precise ideas — form closure and force closure — that tell you exactly when a grip will hold.
Start with form closure. The idea: surround the object with contact points so thoroughly that geometry alone blocks every possible motion. Imagine a key dropped into a key-shaped slot — it cannot rotate or slide in any direction, not because anything is squeezing it, but because solid walls are in the way wherever it might try to go.
The defining feature of form closure is that it needs no friction at all. Even if every contact were perfectly slippery — frictionless ice against frictionless glass — the object still could not move, because the contacts physically box it in. That makes form closure the strongest and most reliable kind of hold, but also the most demanding: fully caging a single rigid body in three dimensions typically takes at least seven well-placed contact points.
Force closure: letting friction do the work
Most real grippers do not cage objects — they pinch them. Two fingertips press on opposite faces of a block and lift it. There are no walls boxing the block in; if the fingers were frictionless, it would shoot out like a wet bar of soap. What saves the grasp is friction, and this is the world of force closure.
Force closure says: a grasp holds if, by squeezing, the fingers can generate contact forces that cancel out any disturbance you could throw at the object. "Any disturbance" includes pushes in every direction and twists about every axis. Roboticists bundle a push-and-twist together into one quantity called a wrench — a force paired with a torque. A grasp has force closure when the contacts can resist every possible wrench.
Friction is what makes this possible, and it has a budget. At each contact the sideways grip force cannot exceed roughly the normal (pressing) force times the friction coefficient — push harder than that and the finger slips. Picture a cone standing on each contact point: as long as the force the finger needs stays inside its friction cone, the contact holds without sliding. Force closure is the statement that, between all the contacts' cones, the fingers can together produce whatever wrench is required.
Two kinds of hold, and how to score them
Lay the two ideas side by side. Form closure traps the object with shape and needs no friction; force closure relies on squeezing and friction to stay put. Any grasp with form closure automatically has force closure too — if geometry already blocks all motion, adding friction can only help. The reverse is not true: a two-finger pinch can have force closure while being nowhere near form closure, because remove the friction and it falls apart.
Knowing a grasp holds is yes-or-no; in practice we also want to know how strongly it holds. That is the job of a grasp quality metric — a single number scoring how robustly a grip resists pushes and twists. A grasp barely scraping by, with the needed force pressed right against the edge of its friction cone, scores low; one with lots of margin to spare scores high.
One common way to compute it imagines the full set of wrenches the contacts can resist as a shape in wrench space — the "grasp wrench set." A widely used metric measures the radius of the largest ball that fits inside that shape: it answers "what is the weakest direction of this grasp, and how big a disturbance can it survive there?" Bigger ball, better grasp. A planner can then compare many candidate grips and pick the best-scoring one.
for each candidate grasp g:
if not has_force_closure(g):
skip # cannot hold at all
score[g] = radius of largest ball
inside g's grasp-wrench-set
choose g with the highest scoreTrade-offs: rigid cage versus friction pinch
Why not always go for form closure, the stronger hold? Because caging an object takes more fingers, more contact points, and a hand shaped to wrap around the part — expensive, bulky, and slow to position. A force-closure pinch needs only two fingers and a moment to close them, which is why parallel-jaw grippers dominate factories and warehouses. The cost is dependence on friction: a slick, oily, or dusty surface shrinks the friction cones, and a grasp that held a dry part may slip on a greasy one.
Real designs blend the two. Fingertips are often given a shallow V-groove or a soft pad: the groove adds a little geometric trapping toward form closure, while the soft pad — a form of compliance — spreads the contact and enlarges the effective friction, strengthening force closure. A tactile sensor in the fingertip can feel the first hint of slip and tell the controller to squeeze a bit harder before the object escapes.
There is also a softer middle ground called caging, which sits just short of full form closure: the fingers loosely surround the object so it cannot escape the enclosure even though it can still rattle around inside. Caging is forgiving when you are unsure of the object's exact pose — you trap it first, then close in to a firm grasp. And sometimes the smartest move is not to grasp at all: non-prehensile manipulation like pushing or nudging a part into a corner can set up an easy, high-quality grasp that would have been impossible from the object's starting position.