LaST-VLA: Thinking in Latent Spatio-Temporal Space for Vision-Language-Action in Autonomous Driving

Open paper
Benchmark Model Rank Results
navsim-on-opensceneLaST-VLA-8B (RL)#15PDMS: 91.3
navsim-on-opensceneLaST-VLA-2B (RL)#18PDMS: 91.1
navsim-on-opensceneLaST-VLA-8B (SFT)#50PDMS: 87.3
navsim-on-opensceneLaST-VLA-2B (SFT)#51PDMS: 87.0

Rank counts only results with a code link.