SlipSense: Multimodal Tactile Learning for Low-Latency and Generalized Slip Detection
SlipSense reports fast slip detection across tactile pressure and vibration signals, including zero-shot transfer to a different robotic hand.
The system pairs a 32 × 32 piezoresistive array with an 8 kHz accelerometer, then fuses the two streams for causal prediction at 240 Hz. In experiments over 1.4 million frames and 37 objects, it reached 96.7% Macro F1 with a false-positive rate under 1.6%. The paper says 76% of slip events were detected within 23.1 ms. Trained only on UMI data, it generalized to a Tesollo dexterous hand without retraining. ArXiv · AI/CL/LG's note
The system pairs a 32 × 32 piezoresistive array with an 8 kHz accelerometer, then fuses the two streams for causal prediction at 240 Hz. In experiments over 1.4 million frames and 37 objects, it reached 96.7% Macro F1 with a false-positive rate under 1.6%. The paper says 76% of slip events were detected within 23.1 ms. Trained only on UMI data, it generalized to a Tesollo dexterous hand without retraining. ArXiv · AI/CL/LG's note
score 4