Lapeyre said selling humanoid robots remains difficult as the use cases are unknown and the systems are still unreliable. Today, the technology is primarily developed by several well-funded companies, including Tesla, Figures, and Agility Robotics. “I want to democratize it by hugging my face,” he says.
Many of the AI models, software frameworks, and tools researchers and engineers rely on to build AI models and applications are already open source. This means that the model will be shared for free and free, along with a license that allows code to be modified and reused. Creating open source hardware usually means that you can release designs, component details, and 3D models to make pieces easier.
The availability of powerful open-weight AI models (which can be downloaded, but not necessarily completely open source) makes it easier for researchers and startups to try cutting-edge AI, and shows how the models work and change their code. According to DeLangue, Face believes it needs something similar to robotics. “Hopefully open source can unlock a wide variety of scope [new robot] Ability,” he says.
Lapeyre adds that open sourcing hardware has similar effects. The robot developer can do it [3D] If something is broken, print the parts,” he says. “If something isn’t perfect, adding new parts can make it a little better.”
The current AI boom is consistent with a renewed interest in robotics, as the latest models help to enable new advances in the functionality of hardware systems. Some well-known researchers argue that AI needs physical presence to fit or outweigh human intelligence.
The hype surrounding the surrounding humanoid robots has led to some questionable claims. Some companies competing to build humanoid robots have posted demo videos on social media that seem to promise incredible capabilities. However, experts warn that such videos can be misleading. Systems that look extraordinarily online can actually be teleospeeded by people away from the camera. A slight change in conditions can cause failure. Or you can’t ensure that the task is completed.
Delangue says an open source approach should make progress more transparent. “You can’t cheat, you can’t hide it in open source,” he explains.
HugFace already hosts some open source robot code. Delangue says that the use of this code has skyrocketed over the past year, reflecting a growing interest in robotics in general.
Some robotics researchers, particularly academia researchers, support an open approach. “Making robotics more accessible will speed up technology moving forward,” says Sergei Levine, an assistant professor at the University of California, Berkeley and co-founder of Physical Intelligence.
Physical Intelligence first made its robot foundation model, PI0, available for use in February. This model allows different robots to learn to perform different physical tasks.
Levine says academia and industry researchers have already contributed valuable ideas and fine-tuning to his products. He added that outsiders could also contribute to the development of new hardware.
“There’s a lot more creativity that people can apply to how they actually build physical hardware,” he says.
The open approach appears to be gaining momentum across the AI industry. Meta was the first major AI company to offer cutting-edge openweight models when it released Llama in 2023. Several other cutting-edge open weight models followed. In January, a relatively unknown Chinese startup called Deepseek shocked the high-tech industry and the stock market by releasing a powerful AI model developed at less cost than companies created by US companies.
Even Openai, a company at the heart of the current boom, has kept its most powerful models in close-guarded secrets, said it will change its approach this summer and release a free, openweight model.
