[论文] SlipSense: Multimodal Tactile Learning for Low-Latency and Generalized...

论文概要

研究领域: ML
作者: Tong Jian, Aditya Thurvas Senthil Kumar, Xinyi Li, Ziling Chen, Tianyu Dai, Ali Sengul, Matteo Grimaldi, Wenjie Lu, Saleh Nabi, Tao Yu
发布时间: 2026-09-14
arXiv: 2609.15910

中文摘要

滑移检测是灵巧操作的基础，然而现有系统往往缺乏对检测延迟和跨平台泛化的精确表征。我们提出 SlipSense，一个基于 TacV5 的多模态触觉滑移检测框架。TacV5 是一个紧凑传感器，集成了以240Hz运行的32×32压阻阵列和以8kHz运行的3轴MEMS加速度计。压阻阵列捕捉空间压力分布，加速度计捕捉摩擦引起的振动，提供互补的滑移线索。该框架执行模态特定编码、传感器内融合和跨模态注意力，并以240Hz进行因果时序预测。在一个包含140万帧、跨越37个物体的数据集上，实验展示了两种模态的互补性。SlipSense 达到96.7%的宏F1，假阳性率低于1.6%，在23.1毫秒内检测到76%的滑移事件。当仅在UMI数据上训练时，SlipSense 零样本泛化到 Tesollo 灵巧手，无需重新训练即可迁移到未见物体、不同传感器单元和机器人平台。

原文摘要

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...

自动采集于 2026-09-16

#论文 #arXiv #ML #小凯