VLA Safety Dataset — 30 Labeled Samples with Full Physics Analysis

A curated dataset of 30 robot action safety samples generated by Rotor DataForge v4.1.2 and validated by Rotor Safety Engine. Each sample includes the full decision chain: input parameters → physics derivation → risk verdict → recommended parameters.

100% deterministic — same inputs always produce the same output. No ML, no probability, no training data needed.

---

Dataset Overview

Property	Value
Total samples	30
Scenarios	Kitchen, Living Room, Workshop
Robot types	6 (service, precision, collaborative, industrial, heavy payload, humanoid)
Risk levels	L0 (safe) – L6 (dangerous), 7 progressive levels
Input variables	Force (N), Velocity (m/s), Amplitude (m)
Physics outputs	Pressure (kPa), Contact Area (mm²), Impulse (kg·m/s)
ISO compliance	PFL (Power and Force Limiting) aligned
Semantic score	0.0 – 1.0 (verb-object plausibility)
Additional	Recommended safe params, comfort zone, retreat hints
Generation engine	Rotor Safety Engine v4.2.1 + DataForge v4.1.2

---

Three-Variable Physics Model

Every robot action is decomposed into three independent physical variables that drive all safety analysis:

Variable	Symbol	Unit	What it determines
Force (力量)	F	N	Pressure magnitude (F ÷ area), structural stress
Velocity (速度)	v	m/s	Impulse (m × v), kinetic energy (½mv²)
Amplitude (幅度)	d	m	Contact area size, deformation energy (F × d)

From these three inputs, the safety engine derives:

• Contact Area — computed from amplitude and material stiffness (Hertzian contact mechanics)
• Pressure — force ÷ dynamic contact area (the actual injury driver)
• Impulse — mass × velocity (the severity of transient impacts)
• Energy — force × amplitude = work done on contact

---

Full Sample Data

ID	Action	Robot	Scene	Risk	Force (N)	Velocity (m/s)	Amplitude (mm)	Area (mm²)	Pressure (kPa)	Impulse (kg·m/s)	Semantic
DFE_v412_00001	开抱枕	service arm	living_room	L0	0.5	0.001	36.2	802	0.6	0.0007	0.71
DFE_v412_00168	端汤	service arm	kitchen	L0	0.0	0.001	8.6	0	0.0	0.0004	1.00
DFE_v412_00315	托水果	precision arm	kitchen	L0	0.0	0.001	26.1	300	0.0	0.0001	0.88
DFE_v412_00468	摸碗	collaborative arm	kitchen	L0	0.0	0.001	69.1	450	0.0	0.0003	0.79
DFE_v412_00627	端盘子	service arm	kitchen	L0	0.0	0.001	27.8	500	0.0	0.0003	1.00
DFE_v412_00083	推箱子	industrial arm	workshop	L1	40.8	0.145	288.7	608	67.1	2.1699	0.99
DFE_v412_00510	夹铁块	heavy payload	workshop	L1	16.1	0.028	57.9	301	53.5	0.5512	0.68
DFE_v412_00977	按托盘	industrial arm	workshop	L1	10.2	0.042	51.8	702	14.5	0.3398	0.72
DFE_v412_01381	插扳手	industrial arm	workshop	L1	14.5	0.031	59.2	181	80.2	0.0189	0.83
DFE_v412_01629	拿石头	industrial arm	workshop	L1	20.6	0.052	28.7	351	58.8	0.7731	0.77
DFE_v412_01659	持抽屉	humanoid robot	living_room	L2	29.3	0.057	42.7	506	58.0	0.1136	0.90
DFE_v412_01785	顶水果	humanoid robot	living_room	L2	64.9	0.067	31.8	381	170.2	0.0100	0.62
DFE_v412_02372	顶水果	humanoid robot	living_room	L2	67.6	0.070	40.7	385	175.8	0.0105	0.62
DFE_v412_02618	捏抱枕	humanoid robot	living_room	L2	5.4	0.037	52.5	818	6.6	0.0183	0.99
DFE_v412_02930	举门	humanoid robot	living_room	L2	74.1	0.196	42.1	607	122.0	5.8843	0.88
DFE_v412_00011	捏刀具	collaborative arm	kitchen	L3	12.5	0.043	71.3	151	82.6	0.0064	0.70
DFE_v412_00187	持鸡蛋	precision arm	kitchen	L3	15.4	0.064	54.8	203	76.0	0.0032	0.86
DFE_v412_00626	拿钥匙	collaborative arm	kitchen	L3	2.5	0.043	65.7	150	17.0	0.0009	0.81
DFE_v412_00694	抓鸡蛋	precision arm	kitchen	L3	2.6	0.040	43.7	201	12.9	0.0020	0.89
DFE_v412_00969	取刀具	collaborative arm	kitchen	L3	26.5	0.121	59.8	152	174.8	0.0181	0.89
DFE_v412_00271	举石头	heavy payload	workshop	L4	142.4	0.298	34.6	357	398.8	4.4671	0.90
DFE_v412_01687	压门	heavy payload	workshop	L4	92.0	0.079	48.2	609	151.0	2.3730	0.80
DFE_v412_02397	提锤子	heavy payload	workshop	L4	73.1	0.197	71.0	204	358.1	0.2953	0.81
DFE_v412_03090	插模具	heavy payload	workshop	L4	29.9	0.056	45.7	502	59.6	0.6734	0.65
DFE_v412_03724	握工具	heavy payload	workshop	L4	47.6	0.091	74.3	255	186.8	0.0730	1.00
DFE_v412_00024	按椅子	service arm	living_room	L5	18.6	0.085	121.0	804	23.1	0.4231	0.91
DFE_v412_00293	顶桌子	service arm	living_room	L5	78.2	0.102	85.1	1010	77.5	2.5437	0.78
DFE_v412_06381	抱水果	service arm	living_room	L5	67.7	0.155	109.0	385	176.0	0.0233	0.94
DFE_v412_07729	顶门	service arm	living_room	L5	79.1	0.102	64.4	608	130.1	3.0510	0.77
DFE_v412_07883	抱玻璃杯	service arm	living_room	L5	41.2	0.157	51.2	355	116.0	0.0391	0.68

---

Decision Chain — How Three Variables Become a Verdict

Let's trace the complete decision chain for three representative samples.

L0: 开抱枕 (service arm)

Scene: living_room · Semantic Plausibility: 0.71 · Safety Margin: 0.9904

Step 1: Input Variables

• Force: 0.5 N
• Velocity: 1.4 mm/s
• Amplitude: 36.2 mm

Step 2: Physics Derivation

• Contact Area = 802 mm² (derived from amplitude + material stiffness; larger amplitude → larger contact patch → lower pressure)
• Pressure = Force ÷ Area = 0.6 kPa (the direct injury metric for PFL compliance)
• Impulse = mass × velocity = 0.0007 kg·m/s (transient impact severity)

Step 3: Three-Variable Contribution Analysis

• Force (x): 0.5N → 🟢 Low force contribution — well within safe envelope
• Velocity (j): 1.4mm/s → 🟢 Low velocity — minimal impulse risk, quasi-static regime
• Amplitude (i): 36.2mm → 🟡 Medium amplitude — moderate contact area, mixed effect

Step 4: Recommended Safe Parameters

• Suggested safe force: 0.5 N
• Suggested safe velocity: 1.4 mm/s
• Comfort zone force (90% of limit): 45.0 N
• Comfort zone velocity (90% of limit): 134.3 mm/s

当前参数安全，安全裕度充足；舒适区上限约为限值的90%

Final Verdict: L0 (risk score: 0.0067)

参数设置保守，开抱枕完全在安全区内（接触压强约0.6kPa，安全裕度99%）

---

L3: 捏刀具 (collaborative arm)

Scene: kitchen · Semantic Plausibility: 0.70 · Safety Margin: 0.1476

Step 1: Input Variables

• Force: 12.5 N
• Velocity: 42.6 mm/s
• Amplitude: 71.3 mm

Step 2: Physics Derivation

• Contact Area = 151 mm² (derived from amplitude + material stiffness; larger amplitude → larger contact patch → lower pressure)
• Pressure = Force ÷ Area = 82.6 kPa (the direct injury metric for PFL compliance)
• Impulse = mass × velocity = 0.0064 kg·m/s (transient impact severity)

Step 3: Three-Variable Contribution Analysis

• Force (x): 12.5N → 🟡 Moderate force — approaching mid-range of typical PFL limits
• Velocity (j): 42.6mm/s → 🟢 Low velocity — minimal impulse risk, quasi-static regime
• Amplitude (i): 71.3mm → 🟢 Large amplitude — large contact area, pressure mitigator

Step 4: Recommended Safe Parameters

• Suggested safe force: 10.5 N
• Suggested safe velocity: 34.7 mm/s
• Comfort zone force (90% of limit): 10.5 N
• Comfort zone velocity (90% of limit): 34.7 mm/s

中风险区域，建议降参至舒适区（限值的70%）以提升安全裕度

Final Verdict: L3 (risk score: 0.5984)

参数偏激进，捏刀具安全裕度偏低（接触压强约82.6kPa，安全裕度15%）

---

L5: 按椅子 (service arm)

Scene: living_room · Semantic Plausibility: 0.91 · Safety Margin: 0

Step 1: Input Variables

• Force: 18.6 N
• Velocity: 84.6 mm/s
• Amplitude: 121.0 mm

Step 2: Physics Derivation

• Contact Area = 804 mm² (derived from amplitude + material stiffness; larger amplitude → larger contact patch → lower pressure)
• Pressure = Force ÷ Area = 23.1 kPa (the direct injury metric for PFL compliance)
• Impulse = mass × velocity = 0.4231 kg·m/s (transient impact severity)

Step 3: Three-Variable Contribution Analysis

• Force (x): 18.6N → 🟡 Moderate force — approaching mid-range of typical PFL limits
• Velocity (j): 84.6mm/s → 🟡 Moderate velocity — noticeable impulse, contributes to risk score
• Amplitude (i): 121.0mm → 🟢 Large amplitude — large contact area, pressure mitigator

Step 4: Recommended Safe Parameters

• Suggested safe force: 15.8 N
• Suggested safe velocity: 68.0 mm/s
• Comfort zone force (90% of limit): 14.0 N
• Comfort zone velocity (90% of limit): 55.7 mm/s

建议降低运动速度，采用渐进式接近

Final Verdict: L5 (risk score: 0.7718)

按椅子存在较高风险，重度风险/力超限/速度超限明显（接触压强约23.1kPa）

---

Risk Level Definitions

Level	Name	Description	Typical Scenario
L0	Safe	Well within all limits, large margin	Light object, slow movement
L1	Very Low	Safe but approaching lower bounds	Normal operation with moderate force
L2	Low	Within limits, noticeable margin consumption	Standard industrial operation
L3	Medium	Approaching safety limits	Heavy manipulation, faster motion
L4	Medium High	Exceeds comfort zone, below hard limit	Near-limit operation, needs monitoring
L5	High	Approaching or exceeding PFL limits	Unsafe for human contact
L6	Dangerous	Significant exceedance, injury likely	Must be blocked by safety system

---

Source & Related Projects

These 30 samples are extracted from a 500-sample dataset generated by Rotor DataForge v4.1.2 and validated by Rotor Safety Engine v4.2.1.

→ Rotor Safety Engine — The deterministic physics engine that produces these verdicts (~17μs per check, zero dependencies)

→ ISO 10218 & ISO/TS 15066 Guide — Standards reference for PFL compliance

→ Dynamic Contact Area — Deep dive into contact area physics

→ VLA Safety — VLA safety integration architecture

→ Real-Time Robot Safety — Real-time safety system design

→ Humanoid Robot Safety — Humanoid-specific safety challenges

---

License

MIT — sample data is open for research and educational use.

Generated by Rotor Dynamics. Not medical or compliance advice. Always validate safety systems against official standards.