Where tactile is necessary

Tactile sensing addresses failure modes that occur when visual perception of the contact interface is incomplete or absent.

Vision-based manipulation systems work well when the camera can see the relevant geometry. They degrade when contact surfaces are hidden, when the task requires force information that geometry cannot provide, or when the control loop needs feedback faster than visual inference can deliver.

These are structural gaps in what vision can measure. The following three application contexts represent tasks where tactile measurement is necessary for reliable execution.

Occluded grasping

Camera visibility is incomplete when the hand closes around an object.

A robot reaches for an object. The vision system estimates pose. As the gripper closes, the hand occludes the object. The contact surface is now hidden from view. The vision system's estimate of contact state becomes stale or uncertain.

Grasping through clutter amplifies this problem. Multiple objects, partial views, and reflective surfaces all degrade visual inference. The gripper must make contact decisions with incomplete visual information.

Force and slip signals from the contact surface provide the necessary measurement. The gripper can determine whether contact has been made, whether the grip is stable, and whether slip is occurring. These signals come directly from the contact interface.

In-hand reorientation

Repositioning an object in-hand requires continuous monitoring of grip force and slip onset.

In-hand reorientation is a manipulation primitive. The robot rotates or translates an object within its grip, repositioning it for the next task step without placing it down.

This motion is dynamic. Contact points shift. Forces redistribute. The object must not slip out of the grip during reorientation. The grip must maintain force closure throughout the motion.

Vision cannot observe the contact interface during this motion. The fingers occlude the object. Force closure is a contact condition measurable only at the surface where force is applied. Tactile sensing provides the continuous feedback needed to adjust grip force and prevent slip during in-hand motion.

Contact-critical force tasks

Tasks requiring calibrated force at a specific contact point.

Some manipulation tasks require applying controlled force at a specific location. Insertions, placements, cap removal, connector mating. The task succeeds or fails based on force magnitude and distribution at the contact point.

Visual feedback cannot tell you whether the applied force is correct. It can estimate contact from geometry, but not the force profile. A camera sees that contact has occurred. It does not measure how hard the surfaces are pressing against each other or how force is distributed across the contact patch.

Tactile sensing measures force directly at the interface. The control loop can regulate applied force based on measurement, not inference. This matters for tasks where the margin between too little force (task fails) and too much force (object damaged) is narrow.

Common thread

These three application contexts share a structure: the task requires information about contact state that visual inference cannot provide reliably. The contact surface is occluded, the needed information is about force distribution, or the control loop requires measurement rather than estimation.

Tactile sensing fills these gaps. The modalities are complementary: vision for pose estimation and task context, tactile for contact state and force information at the interface.