[Orca Hand] Sensor noise filtering and contact label quality for builders researchers (advanced)
Orca Hand discussion on tactile sensor noise, filtering strategies, label quality, and how noisy contact signals affect datasets and policies.
Orca Hand tactile data becomes much less useful when noise and unstable thresholds leak directly into contact labels or downstream grasp-learning datasets.
How are you filtering sensor noise and deciding whether contact labels are reliable enough to keep?
Please share how you separate real contact from drift or vibration, what filtering approach you trust, and how label quality is checked before training or evaluation.
If you reply, include one exact noisy-signal symptom and one exact filtering or QA step that improved label quality.
The best replies will connect filtering choices to measurable label quality or grasp outcomes, not just smoother plots.
If you compare raw vs filtered traces during a grasp event, say what signal change made the labels trustworthy.
Searchers often need to know whether the fix belongs in hardware mounting, filtering, thresholding, or post-label QA.