Lingjun Mao

PhD student at @ucsd_cse | Research in VLM & Multimodal Learning | Formerly @BerkeleyNLP

[Biwei Huang @huang_biwei]

Next AI paradigm: Correlation -> Causal Intelligence

[Lianhui Qin @Lianhuiq]

It’s fun to watch a coding agent reason through spatial construction, iterating through trying, failing, revising, and trying again. Really promising, though still a long way to go.

It reminds me of a kid playing with LEGO for the first time, gradually turning trial and error into something creative, like a piece of art.

Try SimWorld Studio to build your own physical world.

[Lingjun Mao @Lingjun_Mao]

🌍 Coding agents can now automatically build interactive 3D worlds!

🛠️ What’s especially exciting is not just the final result, but the process itself:

🧠 Agents reason about spatial layouts, notice what looks off, replace unsuitable assets, adjust placements, and refine the scene step by step through iterative feedback.

🚀 With SimWorld Studio, agents are no longer just understanding the world, they’re starting to actively create it.

[SimWorld @simworld_ai]

🌊🏝️🌉Coding agent performing spatial reasoning to construct complex scenes

Powered by SimWorld Studio (link in the thread)

[Murray Kang @haoqik322]

🚀 Ready to vibe code physical worlds? Meet SimWorld Studio — an open-source platform for using coding agents to build interactive simulated worlds for embodied agents. 🌍

💡Our goal is to make building interactive physical worlds much more accessible with UE5. You can chat to create environments, place assets, test physics, simulate traffic systems, and edit everything live. 🏙️

This is just the beginning. ✨ We’ll keep updating and expanding the platform, and we’d love for more people to try it out, build with it, and share feedback. 🙌

[Lingjun Mao @Lingjun_Mao]

🚀 With 𝗦𝗶𝗺𝗪𝗼𝗿𝗹𝗱 𝗦𝘁𝘂𝗱𝗶𝗼, you can now vibe code an entire interactive physical world!

💬 Just describe what you want, and the coding agent will build the environment, place assets, and make it physically interactive.

🛠️ In this demo, we directly integrated 𝗖𝗹𝗮𝘂𝗱𝗲 𝗖𝗼𝗱𝗲, equipped it with a diverse set of skills and tools, and used it to quickly build a city scene.

✨ Excited to see everyone build their own worlds with SimWorld Studio!

[SimWorld @simworld_ai]

🚨New Release: SimWorld Studio — Vibe Code the Physical World

Today we open source SimWorld Studio, a coding-agent platform for building interactive physical worlds.

Just chat with Claude Code to create environments, place assets, test physics, and edit everything live.

Build worlds as easily as just writing prompt.

[Lingjun Mao @Lingjun_Mao]

Cool! 🤖 Coding agents can now autonomously build interactive virtual worlds from a simple prompt.

With Unreal MCP tools, the agent can place assets, edit and generate scenes, and even define NPC behaviors to create dynamic gameplay.

Coding agents plus simulators might be the future of interactive game generation. Compared to video generation, it’s much more controllable and supports longer-horizon interactions. 🚀

More demos coming soon!

[SimWorld @simworld_ai]

Claude Code can now build things in a simulated physical world!🤖🏙️

With SimWorld, coding agents can construct buildings, plan cities, or even create video games inside a realistic simulation on Unreal Engine.

Just write a prompt, your agent will call tools, retrieve assets, plan scenes, and test physics autonomously.

Demo platform coming soon so everyone can try it. Stay tuned. 🚀

[Jixuan Chen @chenjx210734]

🚀Excited to share that we bridge the connection of Clawbot & Simworld!

🧩We are motivated to move beyond isolated toy tasks and into a shared physical world with routines, interactions, and coordination.

🚧Lightweight setup: plug in your own agent easily!

[Lingjun Mao @Lingjun_Mao]

Explore everything on our website:

🌐 Website: https://t.co/lYj4acuaxL

📄 Paper: https://t.co/7PYZrrKsb4

💻 Code: https://t.co/fJmiHZNqoT ...9/

[Lingjun Mao @Lingjun_Mao]

🤖 Can an agent earn money by delivering food in a realistic 3D city?

🚚 We present 𝗗𝗲𝗹𝗶𝘃𝗲𝗿𝘆𝗕𝗲𝗻𝗰𝗵, a realistic embodied benchmark for long-horizon food delivery.

⚖️ To earn more, agents must make trade-offs under multiple, interacting constraints (e.g., deadlines, expenses, and battery levels).

😮 Surprisingly, even top models (e.g., Gemini-2.5-Pro, Claude-3.7-Sonnet) earn 𝗳𝗮𝗿 𝗹𝗲𝘀𝘀 per hour than humans. They still make basic mistakes, like packing hot meals together with ice cream.

👇 Project website + more details in the thread ...1/

[Lingjun Mao @Lingjun_Mao]

Despite current agents underperforming humans, the following interventions 𝘀𝘂𝗯𝘀𝘁𝗮𝗻𝘁𝗶𝗮𝗹𝗹𝘆 𝗯𝗼𝗼𝘀𝘁 𝗽𝗿𝗼𝗳𝗶𝘁:

1⃣ 𝗖𝗼𝗻𝘁𝗲𝘅𝘁 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴: asking the agent to write explicit “delivery notes” (e.g., recharge <20%, separate ice cream) raises GPT-5 from $𝟮𝟳.𝟰 → $𝟯𝟲.𝟮/𝗵𝗼𝘂𝗿. 🧠
2⃣ 𝗛𝘂𝗺𝗮𝗻-𝘁𝗿𝗮𝗷𝗲𝗰𝘁𝗼𝗿𝘆 𝗦𝗙𝗧: human demonstrations with action traces + textual annotations (e.g., rationale, reflection, and future plans) turn LLaVA-OneVision-8B from loss-making to $3.2/hour, and improve human-like order batching. 🎯 ...8/

[Lingjun Mao @Lingjun_Mao]

https://t.co/u6YiLT57ZM

Great read! Strongly agree that offline datasets alone don’t translate well to robotics/embodied AI—good embodied agents (and world models) need massive interaction. Real-world data collection is painfully expensive, which is exactly why we need near-realistic, high-freedom simulators to scale interactive learning.