{"id":"3dcfc59a-b14a-4f47-9b97-a4baee2552ae","arxiv_id":"2504.12045","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A new dataset and Gymnasium environment enable detection of pool balls from arbitrary-view photos and simulated shot selection, where a geometry-based Oracle outperforms standard RL agents.","lead":"This paper builds a system that detects pool balls and table markings from a single photo and suggests a good shot. It introduces a dataset and a simulated 8-ball environment, and shows that standard reinforcement-learning agents fail while a geometry-based baseline succeeds in simulation.","discovery_kind":"new_application","skeptic_critique":null,"referee_report":null,"author_rebuttal":null,"desk_editor":null,"rs_alignment":null,"lean_confirmation":null,"pith_extraction":null,"created_at":"2026-08-16T12:40:08.980353+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":null,"supporting_citations":[],"review_version":1}