scripod.com

Building the "App Store" for Robots: Hugging Face's Thomas Wolf on Physical AI

Training Data

2025/09/09
Training Data

Training Data

2025/09/09
In this conversation, Thomas Wolf of Hugging Face discusses the next frontier of open-source AI: robotics. Moving beyond language models, he reveals how Hugging Face is leveraging community collaboration to make physical AI accessible to developers worldwide.
Thomas Wolf outlines Hugging Face's push into robotics through LeRobot, an open ecosystem combining affordable hardware, shared datasets, and developer tools. By lowering entry barriers, the project aims to transform software engineers into roboticists, much like app developers transformed smartphones. A key challenge is data scarcity—unlike text, diverse robotic behavior data is limited—so incentivizing community contributions is critical. Open-source video generation helps simulate training environments, while humanoid robots remain promising but expensive. Wolf emphasizes open science as essential for innovation, transparency, and global participation. He envisions a future where AI empowers creation over consumption, driven by curiosity and open access rather than proprietary control. The discussion also highlights shifting dynamics in the AI world, with China advancing open models and Western companies reevaluating openness, all underscoring the importance of collaborative progress.
03:04
03:04
LeRobot aims to make robotics as accessible as software development.
10:49
10:49
LeRobot is designed as an open platform where startups can build on affordable robotic hardware for new business ideas.
16:07
16:07
Open-source AI can generate realistic, interactive films and training data for robots
29:27
29:27
China is becoming the champion of open-source in AI model development due to hiring advantages and competitive openness.
39:59
39:59
The future of AI lies in enabling everyone to build, not just consume.