Tactile Servoing

The primary data source for tactile servoing. These sensors measure the contact state between the robot and its environment, providing signals like normal force, shear force, and contact geometry. Common technologies include:

• GelSight-type sensors: Use a camera to capture high-resolution deformation of a soft, illuminated gel surface.
• Resistive/ capacitive arrays: Dense grids that measure pressure distribution.
• BioTac sensors: Biomimetic sensors that mimic human skin, measuring pressure, vibration, and temperature. The sensor's spatial resolution, bandwidth, and dynamic range directly determine the fidelity of the servoing control loop.

This software module processes raw tactile signals into a meaningful state representation for the controller. It involves:

• Signal conditioning: Filtering noise and calibrating sensor readings.
• Feature extraction: Deriving task-relevant features from the tactile image or signal. For contour following, this might be the center of pressure or the orientation of a contact edge. For maintaining grip, it could be the incipient slip detection from high-frequency vibrations.
• Sensor fusion: Combining tactile data with proprioceptive (joint encoders, motor current) and exteroceptive (vision) data to form a complete state estimate. This is critical for tasks requiring multi-modal perception.

The core algorithm that closes the feedback loop. It compares the current tactile state to a desired tactile state and computes corrective motor commands. Key controller types include:

• Proportional-Integral-Derivative (PID) control: Simple, widely used for regulating a scalar tactile feature like contact force.
• Impedance/Admittance Control: Regulates the dynamic relationship between motion and force. Impedance control commands force in response to measured motion error, while admittance control commands motion in response to measured force error. Admittance control is often used for compliant tactile servoing.
• Model Predictive Control (MPC): Uses a dynamic model to predict future tactile states and optimize a sequence of control inputs, handling constraints effectively.

The physical hardware that executes the controller's commands. Its performance limits the entire system. Critical characteristics include:

• Backdrivability: The ease with which external forces can move the motor. High backdrivability is essential for force-sensitive and compliant interactions.
• Torque density & bandwidth: The ability to deliver precise torques quickly.
• Compliance mechanisms: Components like Series Elastic Actuators (SEAs) introduce intentional spring-like compliance between the motor and the link, enabling safer, higher-fidelity force control and shock absorption, which is ideal for tactile interaction.

Defines the reference signal for the servo controller. This can be:

• A fixed setpoint: e.g., "maintain 2N of normal force."
• A time-varying trajectory: e.g., "sweep the tactile sensor across the surface while maintaining a constant shear force."
• A policy or skill model: In learned tactile servoing, a neural network policy can directly map the current tactile (and visual) state to desired motor actions or a desired next tactile state, bypassing explicit geometric planning.

The software infrastructure that guarantees deterministic, low-latency execution of the control loop. Tactile servoing often requires loop frequencies of 100-1000 Hz. This framework typically involves:

• A real-time operating system (RTOS) or a real-time kernel patch for Linux.
• Deterministic communication buses (e.g., EtherCAT) for sensor and actuator I/O.
• Hardware-in-the-loop (HIL) simulation tools for safe testing and development before physical deployment. Latency or jitter in this pipeline can destabilize the high-gain feedback control, leading to poor performance or oscillations.

Tactile servoing is critical for peg-in-hole and connector mating tasks where visual occlusion is common. The controller uses force-torque and tactile array feedback to make micro-adjustments, aligning parts despite positional uncertainty.

• Key Mechanism: Maintains a desired contact force profile while searching for the correct alignment.
• Example: Inserting a USB connector or assembling electronic components onto a circuit board where vision alone fails due to tight tolerances.

This application involves guiding a robotic end-effector, such as a probe or tool, along the surface of an object to trace its shape or inspect for defects. Tactile servoing regulates the contact normal force to prevent loss of contact or excessive pressure.

• Key Mechanism: Uses tangential force and slip detection to follow edges and curves.
• Example: Automated non-destructive testing (NDT) for weld inspection or robotic polishing of complex, curved surfaces like turbine blades.

Handling soft, flexible, or fluid-filled objects (e.g., cables, fabrics, food items) is highly challenging for purely vision-based systems. Tactile servoing provides direct feedback on object deformation and slip.

• Key Mechanism: Adjusts grip force and finger positioning in real-time based on tactile shear stress patterns to prevent crushing or dropping.
• Example: Robotic suturing in surgery, folding laundry, or manipulating dough where the object's shape changes during the task.

In shared workspaces, tactile servoing ensures safe and compliant physical interaction. The robot can detect unintended contact and respond with low impedance or guided assistance.

• Key Mechanism: Implements admittance control paradigms where external forces, measured by tactile sensors, are translated into compliant motion.
• Example: A robot co-worker handing off a tool to a human, or providing physical guidance in rehabilitation or assistive robotics.

This involves finely repositioning an object within a multi-fingered hand without releasing it. Tactile servoing provides the high-bandwidth feedback needed for finger gaiting and rolling manipulation.

• Key Mechanism: Controls the center of pressure and internal grasp forces to induce controlled slip or rolling of the object between fingertips.
• Example: Rotating a pen into a writing grip or reorienting a screwdriver within a robotic hand to access a screw at a different angle.

When vision is insufficient (e.g., in dark or occluded environments), robots can use tactile servoing to actively explore an object's properties. The controller sequences exploratory procedures (EPs) like pressing, sliding, and enclosing.

• Key Mechanism: Plans motions to maximize information gain about texture, stiffness, and shape from tactile sensors.
• Example: A search-and-rescue robot identifying objects in debris, or a warehouse robot verifying the contents of an unlabeled box through touch.

A robot control technique that uses real-time visual feedback from a camera to guide the motion of an end-effector towards a desired pose relative to a target. Unlike tactile servoing, which relies on contact force and pressure, visual servoing is primarily concerned with position and orientation derived from image features. It is foundational for tasks like part alignment and tracking, but lacks the direct contact awareness crucial for delicate insertion or contour following.

A robot control strategy that regulates the dynamic relationship between force and motion at the end-effector, making the robot behave as a programmable mass-spring-damper system. It defines a desired impedance (stiffness, damping), allowing the robot to compliantly interact with the environment. Tactile servoing often uses impedance control as its underlying framework, where the tactile sensor data provides the error signal to modulate the robot's compliant behavior in real-time.

A robot control strategy where measured external forces (e.g., from a force-torque sensor) are used to compute a desired motion, effectively making the robot's end-effector move in response to contact. It implements a desired admittance (inverse of impedance). While tactile servoing can employ admittance control, it specifically utilizes high-resolution tactile array data (not just net force) to derive richer spatial contact gradients for more nuanced motion correction.

A high-resolution optical tactile sensor technology. It uses a camera to capture the detailed deformation of a soft, illuminated elastomer gel surface upon contact. GelSight sensors provide dense 3D geometry and texture maps of the contact patch, making them a premier sensing modality for tactile servoing. The rich data enables servoing on fine surface features, such as following the edge of a coin or maintaining uniform pressure across a curved object.

The fine-grained control of an object within a robotic hand's grasp, using coordinated finger motions to reposition or reorient the object without releasing it. Tactile servoing is a core enabling technology for in-hand manipulation, as it provides the continuous feedback needed to adjust finger pressures and positions to prevent slip and achieve precise re-orientation, such as rolling a pen between the fingers.

An advanced motion planning method that optimizes robot trajectories without pre-specifying when or where contacts will occur. The solver discovers contact sequences as part of the optimization to satisfy dynamics and task goals. This is conceptually related to tactile servoing, as both deal with the challenge of planning and controlling through uncertain contact dynamics. Tactile servoing provides the real-time reactive layer that can correct deviations from a planned contact-implicit trajectory.