{"id":"W3126608422","doi":"","title":"A Unified Evaluation Framework for Autonomous Driving Vehicles.","year":2020,"lang":"en","type":"article","venue":"IV","topic":"Autonomous Vehicle Technology and Safety","field":"Engineering","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Huawei Technologies (Canada)","funders":"","keywords":"Reliability (semiconductor); Software deployment; Computer science; Fidelity; Automation; Safety assurance; Test (biology); Reliability engineering; Scenario testing; Engineering; Software engineering; Variety (cybernetics); Artificial intelligence","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":[0.000165789,0.00009950633,0.0001268795,0.00002926489,0.00006824702,0.00001311472,0.0001423205,0.0002001973,0.00006629705],"category_scores_gemma":[0.0001517479,0.0001105011,0.00004805512,0.0001279336,0.00002536462,0.00006401603,0.00002164456,0.0002117584,0.00007925332],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006688179,"about_ca_system_score_gemma":0.00003114638,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":9.840116e-7,"about_ca_topic_score_gemma":0.000002286466,"domain_scores_codex":[0.9994071,0.0000124429,0.0001542438,0.0001527487,0.00008141767,0.0001920187],"domain_scores_gemma":[0.9996483,0.00008828632,0.00002308794,0.0001534882,0.00003158133,0.00005523019],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00006617411,0.0000554361,0.00336526,0.000230191,0.0002791298,0.00001173877,0.003921838,0.1361262,0.01622539,0.1905899,0.005804391,0.6433244],"study_design_scores_gemma":[0.0004165224,0.00005485431,0.003325468,0.00001902528,0.0000397132,0.000001646864,0.00008024959,0.9491607,0.005872089,0.02685783,0.01396203,0.0002098469],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4572651,0.0004512093,0.5315695,0.003866935,0.0003382272,0.0006993561,0.00001017339,0.00253031,0.003269118],"genre_scores_gemma":[0.9871032,0.000008624392,0.01237884,0.000259541,0.0001196777,0.00007541713,0.000008593835,0.0000285158,0.00001759289],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8130345,"threshold_uncertainty_score":0.4506104,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02694590652848825,"score_gpt":0.2580190768332906,"score_spread":0.2310731703048023,"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."}}