{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01090363,0.002035777,0.001150111,0.002982717,0.001049603,0.003296666,0.005950053,0.002074979,0.001623595],"category_scores_gemma":[0.0316451,0.0008101028,0.001593818,0.003562507,0.001003815,0.004554583,0.002821535,0.002567532,0.001113719],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003109375,"about_ca_system_score_gemma":0.00284843,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03855715,"about_ca_topic_score_gemma":0.02964119,"domain_scores_codex":[0.9938282,0.002185211,0.0005992081,0.001270053,0.001758115,0.0003591317],"domain_scores_gemma":[0.9889559,0.003513814,0.0006663306,0.002902117,0.003384942,0.0005769213],"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":[0.0006642496,0.0009593944,0.02797819,0.001032779,0.0005903602,0.0001987105,0.0002989337,0.8007939,0.003853027,0.01730721,0.05096907,0.09535426],"study_design_scores_gemma":[0.00005201455,0.0001805856,0.002891256,0.00008783037,0.00002467438,0.00003991777,0.0001066419,0.9798356,0.003194686,0.006504415,0.007051923,0.00003032767],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.3734213,0.003700847,0.4489045,0.003680886,0.001538976,0.001346937,0.04779453,0.1043541,0.01525786],"genre_scores_gemma":[0.5402857,0.0008703824,0.3147675,0.0005818339,0.0001202913,0.0008559622,0.1378769,0.002986072,0.001655389],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.03855715,"threshold_uncertainty_score":0.0766654,"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."}}