TactileReflex: Noise-Statistics-Driven Vision-Tactile Reflex Control for Force-Sensitive Manipulation

Jul 16, 2026

Upload by Patrick

Manipulating fragile deformable containers requires real-time grip-force adaptation within an extremely narrow force margin where too little force causes slip and too much force irreversibly crushes thin walls. TactileReflex introduces a noise-statistics-based calibration-driven reflex control using vision-based tactile sensing, deriving controller thresholds from intrinsic sensor noise via a brief static-hold-and-unload protocol. A three-channel closed loop uses shear intensity, contact intensity, and center-of-pressure proxies to suppress slip, adapt release, and protect force, achieving 5/5 deformation prevention and 9/10 pouring success.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages8

Reading Time20 min

Introduction

Limitations of existing approaches

Our approach

Instantiating the paradigm

Contributions

TactileReflex: Noise-Statistics-Driven Vision-Tactile Reflex Control for Force-Sensitive Manipulation Ziyan Feng 1 , Yulong Fu 1 , Zheng Li 1 , Yuxin He 1 , Jieji Ren2 , Yudong Zhong 1 , Lujia Wang 1 , Jinni Zhou 1 , Qiang Nie 1,∗ arXiv :2605 .23568v3 [ cs .RO] 11 Jul 2026 Abstract—Manipulating fragile deformable containers, such as disposable plastic cups filled with liquid, demands real-time grip-force adaptation within an extremely narrow force margin: insufficient force causes slip, while excessive force irreversibly deforms the thin wall. Existing approaches struggle to achieve such force-sensitive manipulation tasks. We propose a noisestatistics-based calibration-driven reflex control paradigm with vision-based tactile sensing: by analyzing the sensor’s intrinsic noise characteristics (via a brief static-hold-and-unload protocol), we directly derive all controller thresholds, eliminating external force calibration, trial-and-error manual tuning, or material-specific physical models. Instantiating this paradigm, we present TactileReflex, a three-channel closed-loop controller that extracts three image-level proxies, shear intensity (Sy), contact intensity (Fn), and center of pressure (C), from dual visuo-tactile sensors and drives prioritized reflex channels at ∼12 Hz for slip suppression, weight-adaptive release, and force protection. Each channel closes the loop directly on its proxy via noise-derived thresholds. Ablation demonstrates that only the full three-channel system is able to prevent irreversible container deformation (5/5 success vs. at most 1/5 for partial configurations). In a dynamic pouring task, fixed-effort baselines fail in all 10 attempts due to pose drift, while TactileReflex achieves 9/10 success across two water volumes. As a selfcontained and interpretable controller, TactileReflex can serve as a plug-and-play safety layer beneath high-level manipulation pipelines, including haptic-free VR teleoperation and visionlanguage-action (VLA) policies. Project page: [https://shayfeng.github.io/TactileReflex/](https://shayfeng.github.io/TactileReflex/) I. INTRODUCTION Manipulating deformable containers (soft plastic cups, thin-walled bottles, flexible pouches) is a daily task for humans yet remains challenging for robots. Consider a commonplace disposable plastic cup: it weighs only 3.5–5 g, its wall is just 0.3–0.5 mm thick, it is transparent (offering minimal visual contrast for external cameras), and the margin between the minimum force needed to prevent slip and the force that permanently crushes the cup is remarkably narrow. Now fill it with water and pour: the shifting liquid continuously redistributes the gravitational load along the gripper fingers, demanding real-time effort modulation that fixed-effort or open-loop grasping cannot achieve. The core difficulty is that grasp stability and object safety are tightly coupled: insufficient effort leads to micro-slip and drop, while only slightly more force causes irreversible deformation. A practical grasp controller must therefore (1) detect 1Thrust of Robotics and Autonomous Systems, The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China. 2 School of Mechanical Engineering, Shanghai Jiao Tong University, Shanghai, China. ∗ Corresponding author: Qiang Nie, [qiangnie@hkust-gz.edu.cn](qiangnie@hkust-gz.edu.cn). and suppress incipient slip in real time, (2) reduce effort when the carried load decreases to prevent over-gripping, and (3) enforce a hard safety limit on contact force. The central challenge is ensuring grasp stability while preventing irreversible deformation during dynamic tasks. Limitations of existing approaches. Force/torque sensors can provide continuous feedback for adaptive grasping, but typically require per-sensor calibration and are sensitive to mounting/compliance effects [1], [2] . Vision-based tactile sensors [3], [4] offer rich contact imagery that can address the perception sid

FAQs

What problem does TactileReflex address in force-sensitive manipulation?

It targets grasping and pouring of fragile deformable containers where the safe force range is extremely narrow, so slip can cause drops and slightly higher force permanently deforms the container.

How are controller thresholds determined without external force calibration?

A brief static-hold-and-unload protocol measures the sensor’s intrinsic noise profile, and percentile quantiles (e.g., P95, P99.9) automatically set all controller thresholds.

What signals does TactileReflex use and how do they drive control?

It extracts three interpretable image-level proxies—shear intensity (Sy), contact intensity (Fn), and center of pressure (C)—and closes the loop on each proxy in a prioritized reflex architecture to suppress slip, enable weight-adaptive release, and enforce force protection.

Jul 16, 2026

Upload by Patrick