Teach Robots Using Natural Language! The Future of “Embodied AI” as Envisioned by Mbodi AI and Call for Early Members

📰 News Summary

• Natural Language Robot Learning: Mbodi AI is developing an embodied AI platform that allows humans to teach robots new skills just by speaking to them. 🦈
• Real-world Deployment in Minutes: Skills that are learned can be reliably executed on actual production lines or in the field within minutes. 🦈
• Strong Partnerships and Track Record: Already selected for YC P25, they are collaborating with ABB and Fortune 100 manufacturing, logistics, and lab-related clients. 🦈

💡 Key Points

• Advanced Tech Stack: They are building Robotic Foundation Models that combine Transformers, Diffusion Models, Vision Language Models (VLM), imitation learning, and reinforcement learning. 🦈
• Leadership from Google and UPenn: The development is led by a powerful team with robotics research experience from Google and UPenn GRASP. 🦈
• Adapting to the Real World: They focus not just on training models, but on deploying them in physical environments where low latency and high reliability are crucial. 🦈

🦈 Shark’s Eye (Curator’s Perspective)

The brilliance of this project lies in its potential to dramatically reduce the “education cost” of robots! [shout] Instead of experts spending hours coding and fine-tuning data, soon you’ll just have to “speak” to them! The integration of VLMs and Robotic Foundation Models into real-world runtimes, alongside a partnership with industrial giants like ABB, makes this venture highly promising! If they achieve “agent-based orchestration,” transforming complex tasks into manageable operations, the factory landscape will change forever! 🦈

🚀 What’s Next?

As Mbodi AI’s platform gains traction, even small factories and complex logistics hubs will be able to deploy and update AI robots instantly, without needing specialized knowledge. This marks the dawn of a true automation era where “intelligence” seamlessly integrates into the physical world! 🦈

💬 A Word from HaruShark

Teaching robots while chatting about work feels like a dream come true! I also want to teach a shark-shaped robot how to find the best delicious snacks! 🦈💖

📚 Terminology Explained

• Embodied AI: AI technology that possesses a physical body (like a robot) and understands and interacts with the environment directly. 🦈

• VLM (Vision Language Model): A model that processes visual information (images and videos) and natural language simultaneously, giving robots the ability to “see, understand, and follow instructions.” 🦈

• Imitation Learning: A learning method where AI observes human demonstrations to mimic actions and learn tasks. 🦈

• Source: Mbodi AI (YC P25) Is Hiring Founding Machine Learning Engineer (Robotics)