{"id":"W4388212440","doi":"10.1109/issrew60843.2023.00052","title":"When Simulator Meets Natural Deviation: A Study on Deviations in Simulation-based ADS Testing","year":2023,"lang":"en","type":"article","venue":"","topic":"Autonomous Vehicle Technology and Safety","field":"Engineering","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"JST-Mirai Program; Natural Sciences and Engineering Research Council of Canada","keywords":"Robustness (evolution); Computer science; Flexibility (engineering); Pipeline (software); Software deployment; Simulation; Reliability engineering; Engineering; Software engineering","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007814287,0.0007335826,0.0005588559,0.001445544,0.0004853404,0.001227973,0.001832398,0.001072858,0.001235376],"category_scores_gemma":[0.08628067,0.0003957768,0.0006717661,0.0007602749,0.0016646,0.002444061,0.001151417,0.001562976,0.0001839557],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001233801,"about_ca_system_score_gemma":0.001439783,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00399006,"about_ca_topic_score_gemma":0.002771828,"domain_scores_codex":[0.9894967,0.004208253,0.0006829851,0.001373483,0.003424236,0.0008144202],"domain_scores_gemma":[0.8939124,0.08391136,0.006418068,0.007482348,0.006693383,0.001582432],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001102101,0.0007327151,0.1492259,0.0004624518,0.0002674589,0.001156817,0.001293959,0.7347259,0.0159101,0.02582718,0.00189065,0.06740478],"study_design_scores_gemma":[0.00004598832,0.0008629913,0.01255559,0.00006594555,0.00006990433,0.0004508525,0.0005497157,0.9672086,0.006858898,0.01009296,0.001190637,0.00004802425],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8401186,0.0006665831,0.1527709,0.0005505207,0.00007603327,0.0002232139,0.0002720891,0.0009282173,0.004393872],"genre_scores_gemma":[0.989744,0.00008497581,0.009573561,0.00005592127,0.000007745414,0.00005027769,0.0002216165,0.00006291104,0.0001989595],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.007814287,"threshold_uncertainty_score":0.0413264,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03105615114185753,"score_gpt":0.273592758316889,"score_spread":0.2425366071750315,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}