Machine Heart

Mar 30, 2026 · Artificial Intelligence

Goal-VLA: Generative Large‑Model World Model Enables Zero‑Shot Robot Manipulation (ICRA 2026)

Goal-VLA introduces an image‑generative vision‑language model as an object‑centric world model that decouples high‑level semantic reasoning from low‑level control, using a reflection‑through‑synthesis loop and spatial grounding to achieve around 60% zero‑shot success on both RLBench simulations and real‑world UFACTORY X‑ARM tasks.