SlipSense: Multimodal Tactile Learning for Low-Latency and Generalizable Slip Detection

Forum topic · 小凯 · 2026-09-16

Summary

SlipSense is a multimodal tactile slip-detection framework built on TacV5, a compact sensor combining a 32×32 piezoresistive array running at 240 Hz with a 3-axis MEMS accelerometer sampling at 8 kHz. The piezoresistive array captures spatial pressure distributions while the accelerometer captures friction-induced vibrations, offering complementary slip cues. The framework performs modality-specific encoding, intra-sensor fusion, and cross-modal attention with causal temporal prediction at 240 Hz. Trained and evaluated on a dataset of 1.4 million frames spanning 37 objects, SlipSense achieves 96.7% macro F1 with a false positive rate below 1.6%, detecting 76% of slip events within 23.1 ms. Notably, when trained only on UMI data, SlipSense zero-shot generalizes to the Tesollo dexterous hand and transfers to unseen objects, different sensor units, and different robot platforms without retraining. The work (arXiv:2609.15910) addresses the lack of precise latency and cross-platform generalization characterization in prior slip-detection systems.

Paper Overview

Field: Machine Learning Authors: Tong Jian, Aditya Thurvas Senthil Kumar, Xinyi Li, Ziling Chen, Tianyu Dai, Ali Sengul, Matteo Grimaldi, Wenjie Lu, Saleh Nabi, Tao Yu Published: 2026-09-14 arXiv: 2609.15910

Summary

Slip detection is fundamental to dexterous manipulation, yet existing systems often lack precise characterization of detection latency and cross-platform generalization. The authors present SlipSense, a multimodal tactile slip-detection framework built on TacV5, a compact sensor integrating:

A $32 \times 32$ piezoresistive array operating at 240 Hz, capturing spatial pressure distributions
A 3-axis MEMS accelerometer operating at 8 kHz, capturing friction-induced vibrations

These two modalities provide complementary slip cues. The framework performs modality-specific encoding, intra-sensor fusion, and cross-modal attention with causal temporal prediction at 240 Hz.

Key Results

Evaluated on a dataset of 1.4 million frames spanning 37 objects, demonstrating the complementarity of both modalities
96.7% macro F1 with a false positive rate below 1.6%
76% of slip events detected within 23.1 ms
Trained only on UMI data, SlipSense zero-shot generalizes to the Tesollo dexterous hand, transferring to unseen objects, different sensor units, and different robot platforms without retraining

Original Abstract (excerpt)

> Slip detection is fundamental to dexterous manipulation, yet existing systems often lack precise characterization of detection latency and cross-platform generalization. We present SlipSense, a multimodal tactile slip-detection framework built on TacV5, a compact sensor integrating a $32 \times 32$ piezoresistive array operating at 240 Hz and a 3-axis MEMS accelerometer operating at 8 kHz. The piezoresistive array captures spatial pressure distributions, while the accelerometer captures friction-induced vibrations, providing complementary slip cues. The framework performs modality-specific encoding, intra-sensor fusion, and cross-modal attention with causal temporal prediction at 240 Hz. Experiments on a dataset of 1.4 million frames spanning 37 objects demonstrate the complementarity of t...

--- *Auto-collected on 2026-09-16*

SlipSense: Multimodal Tactile Learning for Low-Latency and Generalizable Slip Detection

Summary

Paper Overview

Summary

Key Results

Original Abstract (excerpt)

Tags