ABot-World Studio - A universal world model workshop launched by Amap (Gaode Maps)
ABot-World Studio is a universal world model workshop launched by AutoNavi, unifying interactive video generation and 3DGS scene generation into a single product. Users can input text or images to generate an AI world that can interact in real time, supporting...
What is ABot-World Studio?
ABot-World Studio is a universal world model workshop launched by AutoNavi, unifying interactive video generation and 3DGS scene generation into a single product. Users can input text or images to generate an AI world that can be interacted with in real time. It supports local deployment of a single 5090 image, and a single continuous inference can exceed 1 hour. It also has a built-in spacetime portal to enable boundless world jumps.
Main features of ABot-World Studio
-
Text/Photos by Life WorldInput text or images to generate an interactive AI world in real time, and output videos and 3DGS files.
-
Real-time interactive explorationSupports WASD and view control, with the world evolving and responding in real time to the user's actions.
-
Anywhere DoorEmbed portals in any scene to jump to another complete 3D world and build a boundless exploration network.
-
Bimodal generationIt simultaneously supports interactive video generation and the generation of 3DGS spatial assets with realistic geometry.
-
Long-term reasoningThere is no upper limit to the inference time at the architecture level, and a single continuous inference can run stably for more than 1 hour.
The technical principles of ABot-World Studio
- ABot-World0 Video Generation ModelThe model integrates scene roaming and character control, and uses algorithm optimization to mitigate error accumulation, enabling continuous reasoning and precise control over longer timeframes.
- ABot-3DWorld0 Scene Generation ModelIt shares the inference path with ABot-Earth0.5, uses the same architecture to connect the generation of indoor, street scene and urban aerial photography at all scales, and adopts a modular design of "generation-evaluation-repair" to ensure high-fidelity output.
- Local Lightweight InferenceThe entire model series is deeply optimized for consumer-grade GPUs and can complete inference on a single RTX 5090, achieving hourly world generation without relying on dedicated computing power clusters.
Follow us on WeChat and reply with "open source",join inAI open source project discussion group
How to use ABot-World Studio
-
Visit the official websiteVisit the ABot-World Studio official website https://abot-world.amap.com/ to access the test page.
-
Input promptsUpload an image or enter a text description to generate the initial world scene.
-
Real-time controlUse the WASD keys on the keyboard and the mouse/arrow keys to navigate in first-person or third-person perspective.
-
Open any doorDiscover or place any door in the scene to travel to a newly generated 3D world.
-
Export assetsSave the generated world as a video file or 3DGS format asset for secondary creation.
ABot-World Studio's core advantages
-
Unified bimodalThis is the first time that interactive video has been integrated with 3D scene generation, balancing real-time exploration with permanent spatial assets.
-
Hourly reasoningA single continuous inference session can last for more than 1 hour without crashes or quality degradation, far exceeding the 1-minute limit for similar systems.
-
Single-card deploymentIt can run on a local RTX 5090, lowering the hardware requirements.
-
Physical consistencyIt highly replicates the rules of real physics, and the world responds precisely to actions, rather than being a passive pixel flow.
-
Full-scale 3DThe same architecture covers indoor scenes, street views, and urban aerial photography, generating scenes with realistic geometry and depth perception.
ABot-World Studio project address
- Project official websitehttps://abot-world.amap.com/
- GitHub repositoryhttps://github.com/amap-cvlab/ABot-World
Comparison of similar products from ABot-World Studio
| Comparison Dimensions | ABot-World Studio | Google Genie 2 |
|---|---|---|
| Reasoning time | A single continuous inference session lasting over 1 hour showed no crashes or quality degradation. | It takes about 1 minute to generate a consistent world, after which it needs to be reset. |
| Modal coverage | It can output interactive video and savable 3DGS spatial assets. | It primarily outputs an interactive 3D video environment, without exporting native 3D assets. |
| Deployment threshold | It can run on a single local consumer-grade GPU (5090) without the need for dedicated computing power. | Currently, it is only in research and preview, and local deployment or external APIs are not yet available. |
| World Expansion | It features a built-in "anywhere door" mechanism, enabling seamless jumps and weaving between multiple worlds. | It possesses long-term memory within a single world, but its ability to expand across worlds is limited. |
| Open source ecosystem | The model and code are fully open source and have been made available on GitHub and Hugging Face. | It is not open source; its capabilities are only demonstrated through the official blog and papers. |
| Application phase | It is now open for public testing and can be used directly for content creation and embodied training. | Positioned as a research tool for rapid prototyping and AI agent evaluation. |
Application scenarios of ABot-World Studio
- Embodied Intelligence TrainingIt provides robots with a high-fidelity physical simulation training ground to simulate perception and decision-making in real-world environments.
- Open World GamesIt supports the boundless dynamic generation of game maps and the arbitrary door jump mechanism, breaking through the traditional level boundary restrictions.
- Film and television storyboard creationIt can quickly convert single-view footage into multi-camera storyboards, reducing the creative verification cycle from weeks to hours.
- Immersive cultural and tourism experienceUsers enter the scene of famous paintings or historical civilizations from a first-person perspective, transforming from passive viewers into first-person participants in the scene.
- Educational training classroomTo create an interactive virtual geography and history teaching space to achieve immersive, inquiry-based learning.