Glossary
Tactile Sensing in Robotics
The use of pressure, force, shear, and vibration sensors embedded in robot hands and grippers to provide physical contact information that vision cannot supply — enabling dexterous manipulation and slip detection.
Tactile sensing in robotics refers to the use of physical contact sensors — arrays that measure pressure, shear force, and vibration across the surface of a robot's end effector — to provide information about contact state, object properties, and incipient failure modes during manipulation.
Vision is the dominant sensing modality in modern robotics, but it has a fundamental limitation: during manipulation, the hand occludes exactly the contact region that matters most. When fingers close around an object, cameras cannot see the contact geometry. Tactile sensing solves this problem by instrumenting the contact surfaces directly.
Types of tactile sensors
Pressure arrays measure normal force distribution across a surface. A dense pressure array can reconstruct the contact geometry — the shape of the contact patch between fingertip and object — which is critical for estimating grasp stability.
Shear sensors measure forces tangential to the contact surface. Shear forces indicate whether an object is about to slip: when the ratio of shear to normal force approaches the friction coefficient, slip is imminent.
Vibration sensors detect high-frequency oscillations caused by surface texture during sliding, and by micro-slip events that precede catastrophic slip. This is the basis of slip detection: Object Class can detect that an object is beginning to slip before it has actually moved, and preemptively tighten its grip.
Thermistor arrays measure thermal properties of contact surfaces, which correlate with material type and help distinguish objects with similar visual appearance but different tactile properties.
How Webbeon implements Tactile Sensing
Object Class integrates 192 taxels per fingertip — each taxel reporting normal force, shear, and vibration at 1 kHz. This dense sensing enables:
- Contact-state reconstruction during occlusion
- Slip detection and preemptive grip correction
- Material classification from contact dynamics
- Force-controlled manipulation: applying consistent forces regardless of object compliance
The tactile signal is processed by a dedicated encoder that produces a compact contact-state embedding, fused with proprioceptive state and visual features in the policy's input representation.
Key facts
- Tactile-enabled Object Class policies converge to stable grasps 3.2x faster than vision-only baselines in training
- Vibration sensing detects micro-slip 40-80 ms before catastrophic slip occurs — enough time for corrective action
- 192 taxels per fingertip at 1 kHz produces approximately 192,000 sensor readings per second per hand
- Tactile sensing significantly reduces the sim-to-real gap because physical contact forces are more directly measurable than visual contact geometry