{"id":"W4388675523","doi":"10.1109/dsa59317.2023.00090","title":"Towards a Benchmark for Trajectory Prediction of Autonomous Vehicles","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":"Ontario Tech University","funders":"","keywords":"Benchmarking; Computer science; Benchmark (surveying); Trajectory; Task (project management); Process (computing); Machine learning; Artificial intelligence; Motion (physics); Data mining; Software; Systems engineering; 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":[0.0001484993,0.00008325658,0.0001352655,0.0001277736,0.00003834754,0.00000269169,0.0001018944,0.0001490364,0.00005099223],"category_scores_gemma":[0.00001503815,0.00008245611,0.00006585469,0.0002035795,0.00004332946,0.00005463819,0.00001628359,0.00008028506,0.00002164637],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003551882,"about_ca_system_score_gemma":0.00002420014,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000006777356,"about_ca_topic_score_gemma":0.00000642869,"domain_scores_codex":[0.9994642,0.000004729823,0.0001931282,0.0001047987,0.00005242952,0.0001806998],"domain_scores_gemma":[0.9997502,0.00004327171,0.00001682987,0.0001447642,0.00002045377,0.00002451072],"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":[0.0001056309,0.0001459802,0.008545302,0.001047941,0.0006301601,0.00001193613,0.002260793,0.08349244,0.14075,0.05608053,0.06483465,0.6420946],"study_design_scores_gemma":[0.001027159,0.0002680733,0.1499562,0.00002890673,0.00005081839,0.000007183779,0.0003711296,0.6793372,0.1339539,0.007838222,0.02684732,0.0003139207],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9566256,0.000139965,0.02143292,0.0001751579,0.0004058525,0.0003541352,0.0001303892,0.004468096,0.01626789],"genre_scores_gemma":[0.9978936,0.0000426707,0.00154168,0.00001054003,0.00003508253,0.00006555983,0.00003026985,0.00001947341,0.0003611503],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6417807,"threshold_uncertainty_score":0.3362463,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01464277401743071,"score_gpt":0.2170985150543173,"score_spread":0.2024557410368866,"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."}}