# Tactus: Open-Vocabulary Object Recognition from Low-Cost Pressure Arrays

arXiv: https://arxiv.org/abs/2608.04043 · HTML: https://mlsanity.com/paper/2608.04043/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 10 of its 14 references that are in the map.
- First posted on arXiv 4 Aug 2026.

## Builds on (most cited first, 10 of 10)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [Masked Autoencoders Are Scalable Vision Learners](https://mlsanity.com/paper/2111.06377/) (2021)
- [Binding Touch to Everything: Learning Unified Multimodal Tactile Representations](https://mlsanity.com/paper/2401.18084/) (2024)
- [A Touch, Vision, and Language Dataset for Multimodal Alignment](https://mlsanity.com/paper/2402.13232/) (2024)
- [Sparsh: Self-supervised touch representations for vision-based tactile sensing](https://mlsanity.com/paper/2410.24090/) (2024)
- [Transferable Tactile Transformers for Representation Learning Across Diverse Sensors and Tasks](https://mlsanity.com/paper/2406.13640/) (2024)
- [TouchFormer: A Robust Transformer-based Framework for Multimodal Material Perception](https://mlsanity.com/paper/2511.19509/) (2025)
- [Jointly Modeling Spatio-Temporal Features of Tactile Signals for Action Classification](https://mlsanity.com/paper/2404.15279/) (2024)
- [Fusion Embedding: A Unified Embedding Space for Text, Image, Video, and Audio](https://mlsanity.com/paper/2607.18666/) (2026)
- [TacVerse: A Multi-Sensor Dataset and Benchmark for Cross-Sensor Vision-Based Tactile Perception](https://mlsanity.com/paper/2606.25877/) (2026)

## Abstract (arXiv)

Resistive pressure arrays are the cheapest and most widely shipped tactile sensors, yet tactile representation learning has concentrated on optical sensors that image a deforming gel. We present Tactus, an open model that answers text queries from pressure data alone: on the STAG benchmark (27 objects, held-out recordings), it reaches 0.771 +/- 0.062 top-1 over four runs (top-3 0.935), matching, and at best exceeding, the dataset's supervised closed-set CNN at 0.76, with no trained classifier head. The recipe is small-data: 187 training recordings, masked-autoencoder pretraining on 144k unlabeled same-sensor frames, and the sensor's own calibration affine, which recovered more accuracy than every architecture change combined. The released model's errors concentrate in a few contact-ambiguous classes, are uncorrelated with text-target geometry (Spearman rho <= 0.05 over 702 class pairs), and survive paraphrased and even bare-name queries within one point; two diverse frames recover 89% of eight-frame accuracy. Failures are reported with equal precision: cross-sensor pretraining pooling gave no gain, vision co-training degraded touch, and a mis-normalized input pipeline silently discarded 97% of the sensor's dynamic range while producing plausible intermediate results. Weights, code, and the memory layer the model plugs into are released openly.