The contact volume must be larger than the object's visual hull

What it is

A teacup handle in the headset is a thin visual rod. Aim the fingers at what you see and the system reports a miss: the selection volume used for intersection is as skinny as the mesh, so tracking error keeps contact flickering on and off. That volume has to outgrow the object's visual hull, so “touching” becomes stable before the hand is geometrically flush with the mesh.

The volume answers whether a grab counts, not how thick the object looks. The mesh can stay thin; the collider can wear its own padding.

Why it happens

Direct touch is a 3D intersection query: a controller tip or fingertip sphere against a collider. Headset hand tracking sits at millimetres to a centimetre or two of steady-state error, and fingers have thickness. If the collider copies a real handle, the query sphere skims past. Noise at the surface then chatters the contact flag at the tracking frequency — a grab that keeps falling in and out.

Padding the collider builds a capture basin: enter the inflated shell and contact latches until the hand has clearly left. The eye still sees the thin handle; the hand meets an invisible shell first. When neighbour spacing drops below twice the pad radius, basins overlap and the wrong object is grabbed — inflation spends spatial resolution to buy tracking tolerance.

Studying it

Run virtual-hand or fingertip-sphere selection while the collider's pad radius relative to the visual mesh is the factor. Thin rods, rings, and keys — stimuli whose visual thickness is near tracking error — expose the volume; cubes hide it.

Independent variables: pad radius, visual thickness, neighbour spacing, injected tracking noise. Dependent variables: time to first contact, contact-flag chatter count, neighbour misgrab rate, rated “did I make contact”.

Drawing the collider as a translucent shell over an opaque mesh shows whether people are catching the shell or the shape. Task time alone conflates a slow grab with a grab that never registered.

Where it stops holding

Large objects (a sofa, a door) gain almost nothing from extra pad, while side-hits still cost. In a dense bundle that must pick one wire, padding must yield to resolution — snap to nearest, or require a second confirm. Force-feedback devices constrain a god-object to the surface, so centimetre-scale padding from a visual-only headset does not transfer. On optical see-through, a volume much larger than the real object makes the hand look as if it is grabbing through air.

Applying it

• Give grabbable objects their own collider, padded one to three centimetres beyond the mesh; thin rods and button rims get more.
• Once contact is on, use hysteresis: a smaller enter threshold than exit, so surface jitter does not chatter.
• When neighbour spacing is below the pad diameter, do not inflate every object equally. Highlight the nearest candidate and confirm, rather than growing all basins.
• How to check: a set of thin handles and rings. Count first-hit rate against neighbour misgrabs. If hits stay flaky, grow the volume; if misgrabs rise, stop growing and change spacing or add a confirm.

Related

• Same group: N2.09.2 The moment of release is harder to judge than the moment of grasp · N2.09.3 A mid-air press has no bottom-out signal and needs another channel · N2.09.4 Vision and sound can fake a touch that is not there
• Nearby: N2.04 Direct Grab · N2.02 Bare-hand Tracking
• Search terms: selection volume · collider padding · capture basin