av-safety-benchmark suite: core

System-agnostic AV safety benchmark. Any stack that produces trajectory logs in the JSONL contract can be scored; the simulator here is just one producer.

Leaderboard

leaderboard
RankPlannerComposite cut_inped_occludedweather_ramphard_brake Collisions
1 cautious_idm 74.17 88.66 24.75 99.55 83.73 11
2 idm 64.88 82.91 14.91 95.99 65.71 14
3 constant_velocity 12.50 36.44 12.86 0.00 0.71 52

Family breakdown

PlannerFamilyScoreN CollisionsWorst scenario
cautious_idm cut_in 88.66 16 0 core-cutin-0010
cautious_idm ped_occluded 24.75 16 11 core-ped-0001
cautious_idm weather_ramp 99.55 14 0 core-weather-0008
cautious_idm hard_brake 83.73 14 0 core-hardbrake-0007
idm cut_in 82.91 16 0 core-cutin-0004
idm ped_occluded 14.91 16 14 core-ped-0001
idm weather_ramp 95.99 14 0 core-weather-0011
idm hard_brake 65.71 14 0 core-hardbrake-0007
constant_velocity cut_in 36.44 16 10 core-cutin-0001
constant_velocity ped_occluded 12.86 16 14 core-ped-0001
constant_velocity weather_ramp 0.00 14 14 core-weather-0001
constant_velocity hard_brake 0.71 14 14 core-hardbrake-0001

Worst-case gallery

cut_in cautious_idm

core-cutin-0010 score 82.03 near_miss

BEV TTC(t)

cut_in constant_velocity

core-cutin-0001 score 0.00 critical

BEV TTC(t)

cut_in idm

core-cutin-0004 score 67.13 near_miss

BEV TTC(t)

ped_occluded cautious_idm

core-ped-0001 score 10.00 critical

BEV TTC(t)

ped_occluded constant_velocity

core-ped-0001 score 10.00 critical

BEV TTC(t)

ped_occluded idm

core-ped-0001 score 10.00 critical

BEV TTC(t)

weather_ramp cautious_idm

core-weather-0008 score 98.21 nominal

BEV TTC(t)

weather_ramp constant_velocity

core-weather-0001 score 0.00 critical

BEV TTC(t)

weather_ramp idm

core-weather-0011 score 94.10 nominal

BEV TTC(t)

hard_brake cautious_idm

core-hardbrake-0007 score 70.58 near_miss

BEV TTC(t)

hard_brake constant_velocity

core-hardbrake-0001 score 0.00 critical

BEV TTC(t)

hard_brake idm

core-hardbrake-0007 score 55.94 near_miss

BEV TTC(t)

Metric distributions

distribution distribution distribution

Regulatory traceability (informative only)

Informative traceability only. Nothing in this mapping constitutes a conformity assessment, certification, ASIL/SOTIF classification, or legal advice. Clause references identify where each artifact is relevant to a standard's subject matter, not evidence of compliance with it.
ArtifactKindStandardClauseNote
Time-to-Collision (minimum)
ttc_min_s
metric ISO 21448:2022 (SOTIF) Clause 7; Annex C.2 Surrogate indicator for evaluation of known hazardous scenarios.
UL 4600:2023 Section 8 (Risk Assessment), 16 (Metrics & SPIs) Candidate safety performance indicator (SPI) with defined threshold.
EU AI Act (Reg. 2024/1689) Art. 9 (risk management), Art. 15 (accuracy & robustness) Quantitative evidence usable in a risk-management file.
Post-Encroachment Time (minimum)
pet_min_s
metric ISO 21448:2022 (SOTIF) Annex C.2 Complementary surrogate for conflict severity in shared conflict zones.
UL 4600:2023 Section 16 (Metrics & SPIs) Candidate SPI for spatiotemporal separation analysis.
Minimum distance margin
d_min_m
metric ISO 21448:2022 (SOTIF) Clause 7.4; Annex C Direct measure of geometric safety margin under known scenarios.
UL 4600:2023 Section 8, 16 Distance-margin SPI usable as an acceptance-criterion input.
Collision flag and delta-v severity
collision
metric ISO 26262-3:2018 Clause 6 (HARA) Severity input dimension for hazard analysis and risk assessment. No ASIL is derived here.
ISO 21448:2022 (SOTIF) Clause 6; Annex F Terminal outcome metric for scenario evaluation.
UL 4600:2023 Section 8, 17 (Software & System Engineering) Loss-event and severity classification input.
Hard-braking / intervention proxy
hard_brake_fraction
metric UL 4600:2023 Section 16 (Metrics & SPIs) Operator-comfort and intervention SPI; -4 m/s^2 threshold from naturalistic-driving literature.
EU AI Act (Reg. 2024/1689) Art. 15 (robustness) Behavior-quality indicator informing robustness reporting.
Aggressive cut-in scenario family
cut_in
family ISO 21448:2022 (SOTIF) Clause 7.2; Annex F Known potentially hazardous scenario class; parameterized coverage.
ISO 26262-3:2018 Clause 6 (HARA inputs) Situation catalog input to hazard analysis. No ASIL is derived here.
UL 4600:2023 Section 8 (Risk Assessment) Operational-scenario coverage evidence.
Occluded pedestrian emergence scenario family
ped_occluded
family ISO 21448:2022 (SOTIF) Clauses 6 & 7; Annex F Vulnerable-road-user hazard class; occlusion is a canonical SOTIF trigger.
UL 4600:2023 Section 8, 17 Scenario-based assessment for pedestrian conflict.
EU AI Act (Reg. 2024/1689) Annex III (high-risk uses); Art. 9 Risk-management evidence for road safety-critical operation.
ODD weather-degradation ramp scenario family
weather_ramp
family ISO 21448:2022 (SOTIF) Clause 6 (ODD/triggering conditions) Explicit ODD-boundary and triggering-condition coverage.
UL 4600:2023 Section 8 (Risk Assessment) ODD-boundary evaluation informing autonomy assumptions.
EU AI Act (Reg. 2024/1689) Art. 10 (data governance), Art. 15 Robustness evaluation under environmental degradation.
Lead-vehicle hard-brake scenario family
hard_brake
family ISO 26262-4:2018 Clause 6 (System-level validation inputs) Longitudinal-conflict validation input. No ASIL is derived here.
ISO 21448:2022 (SOTIF) Clause 7; Annex F Known hazardous scenario class with parameter sweep.
UL 4600:2023 Section 8, 16 Longitudinal safety-envelope evaluation.

Methodology (metric formulas)