{"id":"W4366692929","doi":"10.1007/s10846-023-01808-3","title":"Verifying Collision Risk Estimation using Autonomous Driving Scenarios Derived from a Formal Model","year":2023,"lang":"en","type":"article","venue":"Journal of Intelligent & Robotic Systems","topic":"Autonomous Vehicle Technology and Safety","field":"Engineering","cited_by":7,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Toronto","funders":"Horizon 2020 Framework Programme","keywords":"Computer science; Formal verification; Probabilistic logic; Formal methods; Verification and validation; Collision; Representation (politics); Model checking; Runtime verification; Range (aeronautics); Formal specification; Data mining; Reliability engineering; Artificial intelligence; Software engineering; Algorithm; Programming language; Engineering","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0007749486,0.0002302497,0.0005026847,0.0004034446,0.0002221132,0.00007923672,0.0002984742,0.0002608337,0.000006613263],"category_scores_gemma":[0.0001284254,0.0002150927,0.0001700191,0.0003466967,0.00003633815,0.0004648833,0.00007145627,0.0005327589,0.00005702627],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005117685,"about_ca_system_score_gemma":0.00009896282,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007387785,"about_ca_topic_score_gemma":0.000008761986,"domain_scores_codex":[0.9980137,0.00006363506,0.001065512,0.000152868,0.0003034516,0.0004008466],"domain_scores_gemma":[0.9988649,0.0001768419,0.0004780824,0.000242332,0.0001249927,0.0001128779],"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":[0.00001708807,0.00001832053,0.001046492,0.00004959117,0.0001527684,0.00003956374,0.0007963083,0.9882945,0.004424235,0.00006365577,0.00005218612,0.005045284],"study_design_scores_gemma":[0.0002809693,0.00005714029,0.0007923252,0.0004970684,0.0001065805,0.0001150982,0.0006940037,0.9937324,0.003252113,0.0002433051,0.00002616111,0.0002028239],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4915233,0.0003322649,0.5068921,0.00001356218,0.0008701567,0.0001239002,0.000003746786,0.0002144074,0.00002657895],"genre_scores_gemma":[0.9857071,0.0003577265,0.01369967,0.00000500395,0.0001499896,0.000004713693,0.000006545555,0.00004976519,0.00001946958],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.4941838,"threshold_uncertainty_score":0.8771224,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02529287078887632,"score_gpt":0.2492057805805573,"score_spread":0.223912909791681,"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."}}