LumiTac: Compact, High-Rate Six-Axis Optical Tactile Sensing for Human-Robot Interaction

Hao Wu, Haotian Guo, Yu Feng, Yutong Wang, Jianshu Zhou

November 2026

TL;DR A camera-free optical tactile sensor, 12 mm across and under 2 g, that reads three forces and three moments from 20 colour light channels for about USD 7. A small MLP reaches R² of 0.966-0.994 per axis at about 300 Hz, and the sensor guides a real gripper through admittance control.

What they do

Six-axis force/torque sensing at a fingertip usually means a bulky load cell or a camera-based tactile sensor with limited update rate. LumiTac is a camera-free optical sensor, 12 mm in diameter and 4.75 mm thick, under 2 g and about USD 7 in parts, that estimates three forces and three moments. The authors calibrate it, test tracking, small loads and taps, and use it as a hand-held and gripper-mounted interface for steering a robot.

How it works

Hardware. Two circuit boards joined by translucent silicone. The upper board carries five LEDs (a green one in the centre, red and blue around it); the lower board carries five RGBW photoreceivers, angularly offset from the LEDs. Compression, shear, bending and torsion move the boards relative to each other and change the light each receiver sees.
Readout. The 5 receivers x 4 colours give 20 channels. Finite-element and optical simulation show each load component produces a distinct spatial and colour pattern.
Reconstruction. Channels are baseline-normalized and fed to an MLP (96 and 48 hidden units, 6,966 parameters) trained against a reference six-axis F/T sensor.
Robot use. Forces and moments pass through virtual admittance models to command end-effector translation, rotation and gripper opening.

Results

On an independent test recording, per-axis R² is 0.9657-0.9941 (mean 0.9857). Force MAE is 0.054-0.190 N; moment MAE is 0.304-1.419 mN·m.
Online, the model updates at 303.57 Hz on average with a 0.722 ms median processing latency.
Under 30 s of combined loading, force MAE stays at 0.058-0.240 N. A 10 g weight (about 0.1 N) produces a clear response, and all six rapid finger taps (0.9-3.1 N peaks) are captured.
Demos: an operator drives a simulated Franka through grasp-and-place with the hand-held sensor, and guides a physical gripper’s position, orientation and aperture with the sensor mounted on it.

Limitations

Accuracy is reported on a single held-out recording, and the latency excludes acquisition and transmission. The robot demos are operator-guided motion; the sensor is not yet used as fingertip feedback for autonomous manipulation, and mounting in dexterous hands or legged-robot feet is left to future work.

Figures

From Figure 1: The LumiTac prototype next to a coin, a side view, and concepts for mounting it on a gripper and a dex... — From Figure 1: The LumiTac prototype next to a coin, a side view, and concepts for mounting it on a gripper and a dexterous hand.

From Figure 2: The layered sensor, its LED and receiver arrangement, the 20-channel RGBW readout feeding an MLP, and ... — From Figure 2: The layered sensor, its LED and receiver arrangement, the 20-channel RGBW readout feeding an MLP, and how normal, shear and torsional loads change the internal light paths.

From Figure 4: Calibration against a reference F/T sensor, the preprocessing and MLP pipeline, and test results: per-... — From Figure 4: Calibration against a reference F/T sensor, the preprocessing and MLP pipeline, and test results: per-axis R-squared, error distributions and predicted-versus-reference plots.

From Figure 8: LumiTac mounted on a robot gripper, the admittance control framework, and the robot translation, rotat... — From Figure 8: LumiTac mounted on a robot gripper, the admittance control framework, and the robot translation, rotation and gripper aperture following the applied forces and moments.

Abstract

Integrating six-axis tactile feedback into robotic interfaces requires compact sensors with accurate load measurement and rapid online reconstruction. This paper presents LumiTac, a low-profile, camera-free optoelectronic tactile sensor combining spatially distributed illumination with spectral encoding. Five photoemitters and five RGBW photoreceivers are mounted on opposing circuit boards coupled through translucent silicone. Load-induced interlayer motion modulates the internal light field under compression, shear, bending, and torsion. The sensing head measures 12 mm in diameter and 4.75 mm in thickness, weighs less than 2 g, and has a component cost of approximately USD 7. A compact multilayer perceptron reconstructs three force and three moment components from 20 optical channels. Independent testing yielded coefficients of determination (R-squared) of 0.9657-0.9941, with mean absolute errors of 0.054-0.190 N for forces and 0.304-1.419 mN m for moments. Online prediction achieved an average update rate of 303.57 Hz and a median software processing latency of 0.722 ms. Experiments involving combined loading, light weights, and rapid tapping evaluated load tracking, sensitivity to small forces, and transient response, respectively. Robotic demonstrations further validated its applicability to tactile-guided manipulation and physical human-robot interaction. This study paves the way for compact, high-rate optical tactile sensing in robotic manipulation and human-robot interaction.

Summaries are written automatically from each paper's text. Figures and videos come from the paper and its project page; all credit goes to the authors.

Matt Strong