Tactile Feedback for Contact-Rich Robotic Manipulation

This direction is merged into another direction: visual occlusion in manipulation. Its last version stays readable and citable here.

Version 1 of 137 papers37 conceptsfrozen 2026-10-04

In robotic manipulation, computer vision is excellent at "the first meter"—identifying a target and moving an arm toward it. However, it consistently fails at "the last centimeter." When a robot attempts to plug in a USB cable, thread a needle, or tighten a screw, its own hand inevitably occludes the camera's view. This "last-centimeter" problem is where vision ends and touch begins.

Recent research across dozens of labs suggests that high-precision dexterity requires treating tactile feedback not as a secondary sensor, but as the primary modality for contact-rich tasks. By integrating tactile "skins" and fingertips, robots can navigate environments where they are effectively blind.

Overcoming the Occlusion Barrier

The most fundamental argument for tactile sensing is that vision is physically blocked during the most critical moments of interaction. As Siddhant Haldar, Raunaq Bhirangi, and Lerrel Pinto demonstrate in Touch begins where vision ends, millimeter-level alignment for tasks like key insertion is nearly impossible with vision alone because the robot's end-effector hides the target.

Tactile sensors provide a "ground-truth" for contact that vision cannot replicate. While a vision system might suffer from small calibration errors or pose estimation jitter, a tactile sensor provides immediate, unambiguous confirmation of physical interaction. This allows robots to maintain high-fidelity state information even when the object is entirely hidden within a gripper, a concept explored by Luca Lach and Jan Peters in their work on precise object placing.

Sensing Completion and State Transitions

Touch does more than just locate objects; it detects mechanical states that are visually invisible. In Tactile-Conditioned Diffusion Policy for Force-Aware Robotic Manipulation, Erik Helmut and colleagues show that tactile feedback can signal the "completion state" of a task—such as the specific mechanical resistance felt when a screw is fully tightened or a lid is snapped into place.

Similarly, Xinyue Zhu and Yunzhu Li use tactile signals as triggers for state transitions in delicate tasks like pipette fluid transfer. By detecting subtle pressure changes, the robot knows exactly when to stop or move to the next phase of a sequence, a precision that visual cues often fail to provide.

Emergent Haptic Exploration

When a robot is uncertain about an object's exact position, tactile feedback enables "seeking" behaviors—haptic exploration where the robot feels around to find a feature. Venkatesh Pattabiraman and Lerrel Pinto describe this in Learning Precise, Contact-Rich Manipulation through Uncalibrated Tactile Skins, where a robot haptically explores a surface to locate a slot when its initial visual alignment is off.

This strategy is particularly effective for deformable objects. Neha Sunil and Edward H. Adelson developed a tactile-guided sliding strategy for cloth manipulation. Instead of trying to "see" a hidden corner through layers of fabric, the robot grasps a visible edge and slides its fingers along the material until the tactile sensor detects the specific geometry of the corner.

Manipulation in the Dark

Tactile sensing also serves as a robust alternative in environments where optical sensors fail entirely. Kanishka Ganguly demonstrated zero-shot object grasping using only tactile feedback, successfully picking up transparent and reflective objects that confuse standard depth cameras.

This "vision-free" reliability extends beyond manipulation to navigation. Chaoxiang Ye and Salua Hamaza have shown that whisker-based tactile sensors allow micro-drones to navigate pitch-dark rooms or glass-walled corridors where LiDAR and cameras are useless. By relying on physical contact, these systems bypass the signal loss and noise that plague optical sensors in challenging lighting or material conditions.

Building a Tactile Vocabulary

To make these skills generalizable, researchers are moving toward large-scale tactile datasets and standardized representations. Alymbek Sadybakasov and Jan Peters have worked on building a library of tactile skills that uses raw marker deformations to identify materials and perform in-hand rotation.

The shift is also being fueled by better simulation tools. Shaoxiong Wang and Roberto Calandra introduced TACTO, a simulator that allows researchers to train tactile-aware policies at high speeds. This data-driven approach, combined with "plug-and-play" hardware like AnySkin, is turning the sense of touch from a niche research interest into a standard requirement for the next generation of dexterous robots.

Papers in this version

History & citationVersion 1 · 2026-10-04

Merged into another direction · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_tactile_feedback_for_contact_rich_roboti_e2ee6de8,
  title = {Tactile Feedback for Contact-Rich Robotic Manipulation},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/tactile-feedback-for-contact-rich-robotic-manipulation/rdr_e2ee6de8662443fe8567482cabef6dfc},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_tactile_feedback_for_contact_rich_roboti_e2ee6de8_v1,
  title = {Tactile Feedback for Contact-Rich Robotic Manipulation},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/tactile-feedback-for-contact-rich-robotic-manipulation/rdr_e2ee6de8662443fe8567482cabef6dfc/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}