THE WORLD IS NOT STANDING STILL.RSS
BIG CHANGE.
BIG CHANGE, one tap away
BIG CHANGE, one tap away

Make room for the big picture.

Add BIG CHANGE to your home screen or desktop. It opens in its own window, with the latest stories when you’re online.

iPhone & iPad
Open bigchange.ai in Safari. Tap Share, then Add to Home Screen. Turn on Open as Web App if shown, then tap Add.
Android
Open bigchange.ai in Chrome. In the browser menu, choose Install app or Add to Home screen.
Desktop
In Chrome or Edge, use the install icon in the address bar or the browser’s app menu. In Safari on Mac, choose File → Add to Dock.

Markdown edition

# GelSlim 5.0 switches one fingertip camera between vision and touch

> Paired images reduce depth error on known objects.

By BIGCHANGE.AI Editorial

Published: 2026-10-25T05:06:38.363Z
Updated: 2026-10-25T05:06:38.363Z
Canonical: https://bigchange.ai/blog/gelslim-5-fingertip-camera-vision-touch

![Conceptual charcoal illustration of one flat black fingertip sensor with a circular translucent membrane on its broad face near the rounded tip and an attached amber flex cable.](https://bigchange.ai/api/media/file/gelslim5-broad-face-sensor-hero-v2.png)
AI-generated conceptual illustration by BIG CHANGE, informed by the authors’ sensor figures; not documentary imagery or an independently tested device.

[GelSlim 5.0](https://arxiv.org/html/2610.31334v1) captures UV-off object views and UV-on contact images sequentially under near-stationary contact.

## The big change

- **What changed:** The lab's [GelSlim 4.0](https://www.mmintlab.com/research/gelslim-4-0/) focused on touch; 5.0 adds a visual state.
- **Why it matters:** VisTacFusion combines views for pose and depth.
- **What to watch:** Validation used known objects within sessions.

## What the paired result measures

The [table](https://arxiv.org/html/2610.31334v1) reports depth mean squared error of 0.0066 mm² with paired inputs versus 0.0170 mm² with touch alone, 61.2% lower. One geometry checkpoint served both modes.

Researchers logged about 400 contacts per object across 12 known objects, reserving every tenth session sample for validation and checkpoint selection. Pose queries used object identity; depth targets came from geometry and pose, not membrane measurement. New-object and cross-session results are unreported.

The public [GelSlim 4.0 hardware repository](https://github.com/MMintLab/gelslim_hardware) contains predecessor design files. It does not establish a version 5.0 release. A version 5.0 project page did not resolve during our check; its code, data, hardware files, access terms and cost remain unverified.

## Sources

- [GelSlim 5.0: Unifying Vision and Touch with a Single Fingertip Camera](https://arxiv.org/html/2610.31334v1) — Author paper Sections III to V and Tables II and III establish the prototype construction, VisTacFusion, validation design and author-reported errors. It does not demonstrate deployed manipulation, unseen-object accuracy or independent replication.
- [GelSlim 4.0: Focusing on Touch and Reproducibility](https://www.mmintlab.com/research/gelslim-4-0/) — The lab describes the prior tactile sensor and links manufacturing, design, depth and shear resources. This does not establish GelSlim 5.0 availability.
- [GelSlim 4.0 Hardware Files](https://github.com/MMintLab/gelslim_hardware) — Contains version 4.0 sensor body, assembly, gel mold and electronics file links. It is predecessor context, not a version 5.0 release.