DriveReferee: Geometric Safety Verdicts Need Not Be Learned for Driving World-Action Models

Benchmark Model Rank Results
navsim-on-opensceneDriveReferee (base, 16B backbone)–PDMS: 91.08
navsim-on-opensceneDriveReferee (base, 4B backbone)–PDMS: 90.47
navsim-on-opensceneDriveReferee (full, 16B backbone)–PDMS: 92.02
navsim-on-opensceneDriveReferee (full, 4B backbone)–PDMS: 91.50