A wearable input/output interface learns personalized models of haptic perception

Wen Li, E. Magnus Adams, Zijie Xu, Jiawen Yao, Aybala Esmer, Xiaoyue Ni, Bozhi Tian, Pedro Lopes

November 2026

TL;DR A forearm sleeve pairs a 36-electrode electrotactile array with a capacitive touch layer, so wearers draw where they felt each stimulus and the device learns a personal perceptual map. Across 24 participants, five calibration trials plus a tactile map predict where heat is felt to within 21.41 mm on average, versus 26.65 mm without the tactile map.

What they do

Wearable haptics usually assume a stimulus is felt where the actuator sits, but the same electrode pair can feel different from one person to the next. This sleeve is both an output and an input device: it stimulates the forearm, and the wearer draws on its surface where the sensation was felt. These reports are registered in the electrodes’ own frame and used to fit per-user models that choose which electrodes to fire for a desired sensation.

How it works

Hardware. A 6×6 electrotactile array under a capacitive touch array, plus a Peltier element at the wrist. Electrotactile stimulation makes the Peltier’s warmth feel as if it comes from the tactile site (thermal referral).
Mapping study. 24 participants each felt 28 electrode pairs, alone and combined with heat, and drew each tactile and thermal percept on the sleeve.
Tactile model. The population-mean percept map, shifted by a per-user offset fit from five calibration trials.
Thermal model. A Bayesian model weights other participants’ thermal maps by similarity over five calibration trials. When a user’s tactile and thermal maps align, it also uses their tactile map.

Results

The touch layer localizes touches with a mean error of 1.90 mm on the forearm. Predicting percepts from electrode positions errs by more than 30 mm.
Tactile prediction: 19.35 mm with the offset vs 20.63 mm for the population mean. Thermal: 26.65 mm Bayesian, 21.41 mm with tactile guidance, 14.22 mm for strongly aligned users.
Choosing electrodes for a target thermal location: 19.2 mm error with tactile guidance vs 27.7 mm for the electrode midpoint.
In a new group of 12 participants, model-predicted thermal animations were picked more often than midpoint-based ones.

Limitations

Tactile guidance requires a full tactile map per user and helps little for intermediately aligned users (23.76 to 22.37 mm). Without it, the thermal model does not significantly beat the midpoint at choosing electrodes (25.3 vs 27.7 mm). The user studies used an external stimulator rather than the onboard one, and the four applications are proof-of-concept demonstrations.

Figures

From Figure 1: The haptic I/O sleeve stacks an electrotactile electrode layer and a capacitive touch layer in silicon... — From Figure 1: The haptic I/O sleeve stacks an electrotactile electrode layer and a capacitive touch layer in silicone, with a Peltier element at the wrist. Thermal images show the skin warms only near the Peltier, while the warmth is felt at the electrotactile site. Below, the sleeve on a forearm and its electronics.

From Figure 5: Five-trial tactile calibration with a per-user offset, thermal calibration that sorts users by thermal... — From Figure 5: Five-trial tactile calibration with a per-user offset, thermal calibration that sorts users by thermal-tactile alignment before a Bayesian model, and the prediction and inverse-rendering errors of the electrode midpoint, thermal Bayesian and tactile-guided Bayesian methods.

From Figure 6: The sequential thermal animation study (N = 12) and four demonstrations: in-context calibration in VR,... — From Figure 6: The sequential thermal animation study (N = 12) and four demonstrations: in-context calibration in VR, interpersonal haptic communication, thermal-camera sensory substitution and haptic-guided teleoperation of a fire-rescue robot.

Abstract

Wearable haptic interfaces deliver physical stimuli to the skin, yet perceived locations vary widely across users and diverge from actuator locations, particularly for emergent tactile-feedback techniques. We developed a wearable haptic input/output (I/O) device integrating a 36-channel electrotactile array with capacitive touch sensing, enabling stimulation and user-reported percept locations to be registered within the same body-referenced coordinate frame. In 24 participants, the device revealed electrotactile and referred-thermal perceptual maps with notable interparticipant variability. Despite this variability, both modalities retained electrode-pair-specific spatial structure across participants. Tactile and thermal percepts also showed spatial correspondence, whose strength varied across participants. These findings motivated our human-in-the-loop calibration framework, which adapts a forward-prediction strategy to individual users and generates personalized perceptual models. These models enabled inverse rendering by selecting stimulation patterns predicted to generate a target percept for a specific user. Using these models, we demonstrate personalized haptic rendering in haptic-guided robot teleoperation, visual-to-haptic sensory substitution, interpersonal haptic communication, and virtual reality (VR). Together, this platform provides a route toward adaptive haptic interfaces that measure, model, and render according to each user's perception.

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