{"id":"W3120385278","doi":"10.1109/iv47402.2020.9304698","title":"A Unified Evaluation Framework for Autonomous Driving Vehicles","year":2020,"lang":"en","type":"article","venue":"","topic":"Autonomous Vehicle Technology and Safety","field":"Engineering","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"Huawei Technologies (Canada)","funders":"","keywords":"Computer science; Reliability (semiconductor); Software deployment; Fidelity; Automation; Safety assurance; Scenario testing; Reliability engineering; Track (disk drive); Test (biology); Engineering; Software engineering; Telecommunications; Artificial intelligence; Variety (cybernetics)","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":[0.0001711892,0.000106202,0.000130437,0.00003155999,0.0000719403,0.00001460125,0.0001398596,0.000198344,0.0001448517],"category_scores_gemma":[0.0001472649,0.000108994,0.00004969184,0.0001304223,0.00002433142,0.0000755049,0.00002240976,0.0001805018,0.00006627505],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005341001,"about_ca_system_score_gemma":0.00002765266,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000001103562,"about_ca_topic_score_gemma":0.000003393365,"domain_scores_codex":[0.9993885,0.00001177649,0.0001681796,0.0001570559,0.00008083673,0.0001937001],"domain_scores_gemma":[0.9996337,0.0001113587,0.00001926018,0.0001418862,0.0000353506,0.00005843079],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00003897021,0.00003685062,0.002165154,0.0001446821,0.0002105497,0.00000368087,0.001840457,0.1003389,0.0112012,0.4586788,0.004023442,0.4213173],"study_design_scores_gemma":[0.000300823,0.00005006409,0.001772062,0.00001003747,0.00003050927,0.000001089132,0.0001306198,0.9518092,0.01011555,0.02998228,0.005615456,0.0001822554],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1988699,0.0001812242,0.7890756,0.003064799,0.000156067,0.0005162844,0.000004192255,0.002615609,0.005516324],"genre_scores_gemma":[0.9621679,0.000008598836,0.03727772,0.0003231908,0.00008322562,0.00007776488,0.000007629514,0.00002604752,0.00002794746],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8514704,"threshold_uncertainty_score":0.4444648,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02838895735510281,"score_gpt":0.2586327546624692,"score_spread":0.2302437973073663,"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."}}