Description
This AI Teaches Robots How To Do Chores
🔗 https://github.com/Robbyant/lingbot-video
👉 Latest AI News, AI Tools and AI Prompts: https://franklineh.com
🔥 Join the newsletter free. No spam. Ever. https://franklineh.com/newsletter The complete walkthrough is on youtube.com. What if video generation wasn't just for entertainment, but was actually the "brain" for physical robots? In this video, we're diving deep into Lingbot Video, a groundbreaking open-source project from Robiant (part of the Ant Group). Unlike traditional video AI, Lingbot is designed to simulate the real-world laws of physics, creating a foundation model specifically for robotic training. We break down the brilliant math behind its "Mixture of Experts" (MoE) architecture, how it uses 70,000 hours of embodied data to understand gravity and momentum, and how it's outperforming competitors in physical realism. In this video, you'll learn:
- How Lingbot uses a 30B parameter MoE model to run 3x faster than traditional dense models.
- The secret behind "Embodied Data Grounding" and why it matters for robotics.
- A look at the first-person simulations for domestic tasks like cleaning and making coffee.
- How you can access the open-source weights and code right now. ━━━━━━━━━━ 🔗 SHOW LINKS ━━━━━━━━━━
🔗 https://huggingface.co/collections/robbyant/lingbot-video
🔗 https://github.com/Robbyant/lingbot-video
🔗 https://technology.robbyant.com/lingbot-video ━━━━━━━━━━ 📚 CHAPTERS ━━━━━━━━━━
0:00 - The New Race in AI Video: Teaching Physics
0:24 - Lingbot Video
0:52 - Embodied Video Foundation Model
1:16 - The Architecture
2:22 - Embodied Data
3:02 - Six Signal Rewards
3:31 - Performance Benchmarks
4:18 - Limitations:
4:53 - Real-World Examples
5:54 - Conclusion
Primary source: youtube.com ↗