{"id":"W3206540493","doi":"10.1109/icra48506.2021.9561333","title":"Shaping Rewards for Reinforcement Learning with Imperfect Demonstrations using Generative Models","year":2021,"lang":"en","type":"article","venue":"","topic":"Reinforcement Learning in Robotics","field":"Computer Science","cited_by":13,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University; Vector Institute; University of Toronto","funders":"","keywords":"Reinforcement learning; Computer science; Generative grammar; Artificial intelligence; Machine learning; Generative model; Adversarial system; Function (biology); Action (physics); State space; Convergence (economics); Bellman equation; Range (aeronautics); Imperfect; Imitation; Mathematical optimization; Engineering; Mathematics","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":[0.000291809,0.0001871752,0.0001928958,0.00008703612,0.000568294,0.00041266,0.00030644,0.00005955324,0.00002851841],"category_scores_gemma":[0.00008334201,0.0001640927,0.00008145755,0.0003960778,0.00003735663,0.000954569,0.0001888784,0.000190383,0.000004638233],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001304118,"about_ca_system_score_gemma":0.0004736368,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001805013,"about_ca_topic_score_gemma":0.000009936532,"domain_scores_codex":[0.9984435,0.00007160091,0.000312486,0.0004297332,0.0003489718,0.0003936958],"domain_scores_gemma":[0.9988406,0.0001328458,0.0001380996,0.0003953562,0.0004006288,0.00009251165],"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":[0.000003894955,0.000006313815,0.00007680817,0.00001306231,0.00004767883,0.000006132126,0.0005674257,0.9075794,0.003042993,0.08811917,0.00004211959,0.0004949893],"study_design_scores_gemma":[0.0004135545,0.00020157,0.000007349805,0.00004729118,0.00002148566,0.00003816832,0.0002211962,0.9863222,0.01174856,0.0004317866,0.0003036351,0.0002432002],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.002127425,0.00004377434,0.9884417,0.0002837457,0.0001110057,0.0003059186,2.833866e-7,0.0001915243,0.00849461],"genre_scores_gemma":[0.4707425,0.00001028696,0.5267879,0.0002520333,0.00004905038,0.0000261194,0.00001146021,0.00001460692,0.002106005],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.4686151,"threshold_uncertainty_score":0.6691508,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07543834377601084,"score_gpt":0.2929367701417165,"score_spread":0.2174984263657057,"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."}}