GWM-VLA: Geometry-Aware Latent World Modeling for Vision-Language-Action Learning | Startup Networx | Startup Networx