arXiv:2608. 15265v1 Announce Type: new Abstract: Constructing an interactive 3D open world from a user query is important.
Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv AI.
arXiv:2606. 01057v1 Announce Type: cross Abstract: Procedural 3D modeling through code is emerging as a versatile paradigm, offering deterministic, engine-ready, and precisely editable assets that neural 3D generators inherently lack.
arXiv:2607. 05465v1 Announce Type: cross Abstract: Complex image creation and editing often require more than a single generation or editing model.
Building capable embodied agents requires not only multimodal perception and understanding, but also agentic capabilities for reasoning about actions, adapting to evolving situations, and interacting with the physical world. In this report, we introduce Hy-Embodied-VLM-1.
The paper surveys how large multimodal models (LMMs) enhance agentic frameworks that combine perception, memory, reasoning, planning, and action. It examines the integration of multiple modalities—text, images, audio, and video—through delegated, late‑fusion, and early‑fusion architectures, and maps these designs to agent capabilities. The survey also reviews multimodal agentic systems in robotics, web navigation, multimedia content creation, and video understanding, evaluating performance, efficiency, and scalability trade‑offs.
arXiv:2609.40325v1 Announce Type: new Abstract: As interactive 3D worlds are increasingly used to study intelligent behavior, it becomes important to develop efficient pipelines for identifying anoma...
RoboPhys-3D is a 3D‑grounded embodied world model benchmark built on RoboTwin 2.0, featuring 50 manipulation tasks, 5,000 episodes, and 25,000 multi‑view ground‑truth videos. It evaluates video world models by processing both generated and ground‑truth videos through the same 3D reconstruction pipeline, allowing the separation of reconstruction‑induced from generation‑induced errors. The benchmark defines 50 metrics across four sub‑dimensions—pixel fidelity, 3D geometry consistency, state understanding, and task completeness—and introduces the Average Full Score and RoboPhyscore for holistic assessment, with RoboPhyscore showing strong correlation with human judgments.