{"id":"W4417283698","doi":"10.1145/3748636.3760462","title":"Cognitive Foundation Agents for Generalizable Vision-and-Language Navigation","year":2025,"lang":"","type":"article","venue":"","topic":"Multimodal Machine Learning Applications","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary","funders":"","keywords":"Embodied cognition; Task (project management); Cognition; Process (computing); Adaptation (eye); Asynchronous communication; Key (lock); Suite","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":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0007068318,0.0002517124,0.0002290212,0.0002086618,0.0009078366,0.0006652825,0.0004096969,0.0001588331,0.0001491352],"category_scores_gemma":[0.0004604814,0.0002530855,0.00008990778,0.0007049124,0.0001021059,0.0005733743,0.0003452452,0.0002652188,0.0001269436],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009149624,"about_ca_system_score_gemma":0.0002049918,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001366406,"about_ca_topic_score_gemma":0.00002829579,"domain_scores_codex":[0.9979366,0.000190795,0.000441309,0.0008417119,0.0002517653,0.0003377942],"domain_scores_gemma":[0.9980304,0.0008174891,0.0001936907,0.000436757,0.0004165936,0.0001051302],"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":[0.00004681975,0.000281926,0.002689632,0.0003145814,0.0001250458,0.000001191314,0.005681695,0.001485653,0.002133373,0.2717978,0.001310454,0.7141318],"study_design_scores_gemma":[0.001195681,0.0001104781,0.01988905,0.0002883866,0.00007919178,0.000002375604,0.0002147333,0.9674883,0.00147803,0.00503486,0.003967239,0.0002516489],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.08960913,0.0002166528,0.8981264,0.005459715,0.0004228472,0.001469442,0.00001556861,0.0001429429,0.004537289],"genre_scores_gemma":[0.9116375,0.00003604951,0.07529927,0.001520153,0.00008591867,0.0003374742,0.0001936708,0.00001565491,0.01087427],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9660027,"threshold_uncertainty_score":0.9999921,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01816397981269239,"score_gpt":0.3740435197282998,"score_spread":0.3558795399156074,"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."}}