See, Symbolize, Act: Grounding VLMs with Spatial Representations for Better Gameplay | ScienceToStartup | ScienceToStartup