SlipSense: multimodal tactile slip detection hits 96.7% F1 at 23.1ms with zero-shot robot transfer
chris_j_paxton · x · 2026-10-15
Analog Devices' Dexterous AI Group presents SlipSense, a multimodal tactile slip detection framework built on the TacV5 sensor (32×32 piezoresistive array at 240 Hz + 3-axis MEMS accelerometer at 8 kHz). Using modality-specific encoding, intra-sensor fusion, and cross-modal attention with causal temporal prediction, it achieves 96.7% Macro F1 with <1.6% false positives on a 1.4M-frame, 37-object dataset, detecting 76% of slips within 23.1ms—and transfers zero-shot from a UMI gripper to a Tesollo hand without retraining. Companion work: Touch2Trace, tactile-driven imitation learning for dexterous cable tracing. Podcast episode forthcoming on RoboPapers.
More from Embodied
- Work photos are AI's first mile: if capture is wrong, no model can recover — arnabdotorg · 2026-10-15
- LEGO builds an omnidirectional vehicle for automated warehouses — glenbeer · 2026-10-15
- BlackBerry launches QNX Launchpad to court early-stage Physical AI startups — pdamodaran · 2026-10-15
- Microduck robot runs onboard SLAM with just 64 pixels of depth, 4.8 cm median error — chris_j_paxton · 2026-10-15
- Dynamic Creatures exits stealth as Boston Dynamics' official entertainment and hospitality partner — PeterBowdenLive · 2026-10-15
- Figure AI valued at $39B tops humanoid chart, exceeds any listed robot maker — LexSokolin · 2026-10-15