Lab 15 / 15 · Interactive · real-time
Slip Precursor Detection Bench
Tune the threshold and trade detection rate against false alarms — the precursor lives in the amplitude, not the derivative.
Detector
What to look at
Left plot, ROC: blue = ratio detector (hugs the top-left corner), orange = rate-of-change (hugs the diagonal). Right plot: sample episodes and the threshold line. Switch detectors and watch the whole ROC collapse.
Toy framing: the 80 ms pre-slip window, 1 kHz sampling, and noise σ=0.01 are all settings; what doesn't change is the structure — preslip is a monotonic climb, separable in amplitude, not in derivative.
What problem this lab solves
The key to grip-force control is not «feeling the slip» — it is seeing it 80 milliseconds early in the climb of |f_t|/f_n (T15). This bench brings T15's 2000-trial experiment into the browser: two detectors, one ROC, one threshold knob — whether a false positive crushes an egg or a miss drops the cup, you set the price.
How to read the figure
Left is the ROC: false-alarm rate on the horizontal axis, detection rate vertical, the blue line is the ratio detector (max of |f_t|/f_n in a window), the orange line is the rate-of-change detector (max of dr/dt). The dot is the current threshold. Right is one sample trial: r(t) climbing, noise, the slip moment in red — you can see at a glance where the threshold line sits and when the alarm fires.
Hands-on
Pick the «ratio» detector and sweep the threshold from low to high: detection rate falls, false alarms fall too — find the operating point you can live with (say 95% detection / 1% false alarms). Switch to «rate of change»: the whole ROC collapses toward the diagonal (AUC 0.55, near a coin flip) — that is the hard evidence for «the precursor is in the amplitude, not the derivative».
Done playing — where next