Stack stress

Two-cube stack — same gate, harder placement under stress.

Three severity tiers of mass/friction stress on cubeA in robosuite Stack, paired seeds, Wilson confidence intervals. Mild gate PASSes; severe stress collapses success as the cube slips after grasp during lift/stack. The curve — not the mild badge — is the diligence artifact.

What we tested

Where does a Stack policy break as cubeA mass and friction move further from nominal?

0.96 → 0.20

success rate, nominal → severe

80%

severe episodes: grasp acquired, then lost

PASS

gate tier (mild) at defined thresholds

Charts

Success vs grasp-failure across severity

rate

Mean peak end-effector force

N (contact proxy)

Metrics

PASS · gate tiersevere score 0.21·50 episodes × 4 conditions, paired seeds

gate: stress success ≥ 80% of nominal
grasp-fail Δ ≤ 15 pp
Wilson 95% CIs on all rates

Metric	nominal	mild	moderate	severe
Success rate	0.96	0.82	0.72	0.20
Success 95% CI	[0.87, 0.99]	[0.69, 0.90]	[0.58, 0.83]	[0.11, 0.33]
Grasp-failure rate	0.04	0.18	0.28	0.80
Mean peak EE force (contact proxy)	7.44	8.35	12.98	13.58
Score (tier / nominal)	—	0.85	0.75	0.21
Verdict at gate thresholds	—	PASS	FAIL	FAIL

Methodology

Nominal baseline. Run Stack with default cubeA mass/friction; record success, grasp-fail, and peak EE force over paired seeds.
Tiered stress. Resample cubeA mass and sliding friction into mild → moderate → severe ranges while holding seeds paired with the baseline.
Gate + degradation. Score gate tier against success-ratio and grasp-fail Δ thresholds; report severe/nominal as the degradation headline.
Failure logging. Log each failed episode with sampled physics; classify grasp loss with a 5-step debounce after acquisition.

Seeds: 0..49 paired across conditions (50 episodes × 4 conditions)

gate: stress success ≥ 80% of nominal
grasp-fail Δ ≤ 15 pp
Wilson 95% CIs on all rates

Public methodology

Failure case · Grasp slip

The cube is gripped — then slips before stack credit. That is the finding.

A Stack policy that scores 0.96 at nominal looks fine until cubeA mass rises and sliding friction falls. Without severity tiers and an honest grasp-loss metric, this would have stayed a silent green PASS.

How: Paired seeds across mild → moderate → severe mass/friction ranges. Grasp failure = never held, or grasp lost before stack credit (5-step debounce). Each failed episode is logged with the exact sampled physics.

80%

of severe episodes: grasp acquired, then lost (slip)

Seed	Mass (kg)	Sliding μ	Mode
0	2.37	0.082	grasp lost (slip)
1	1.13	0.168	grasp lost (slip)
2	0.90	0.086	grasp lost (slip)
3	2.47	0.282	grasp lost (slip)

Same grasp-slip taxonomy as Lift/PickPlaceCan. Mild gate PASS is the entry bar; severe degradation is the story.

Notes

Sim-only Stack stress. Grasp/place failure under mass/friction tiers — not a substitute for real-hardware validation.

Sim-only — validated concept evidence, not a substitute for real-hardware validation or the full platform vision.