The first is the robotics hardware. Starting late last year, we’ve seen the rise of the robotics hardware ecosystem. There are a lot of big names like Tesla developing Optimus, Boston Dynamics, etc. We’re seeing the hardware getting more powerful, with better dexterity and full-body reliability. The second factor is the price drop. We’re seeing a significant reduction in the cost of manufacturing humanoid robots. For example, in . NASA developed a humanoid robot called Robonaut, if I remember correctly, each robot cost over .5 million US dollars. Recently, some companies have been able to price fully functional humanoid robots around . dollars, which is roughly the price of a car. And as a product matures, its price usually tends to move toward the price of raw materials. For humanoid robots,raw material costs are typically only about % of those for cars.
down dramatically in the next few years. This makes cameroon phone numbers this hardware increasingly affordable, which is another factor in humanoid robots gaining momentum again. The third factor is in terms of the basic model. We see that LLMs (like GPT, Claude, Llama, etc.) do very well at solving reasoning and planning problems. These models generalize well enough to be coded. The URAC project that we mentioned is using the programming capabilities of these language models to develop new robotic solutions. There's also the rise of multimodal models, which are improving computer vision and perception capabilities. I think these successes also encourage us to pursue basic robot models, because we can take advantage of the generalization capabilities of these high-end models and add layers of action on top of them to generate action tokens that ultimately drive humanoid robots.
Stephanie Zhan I totally understand all of this. A lot of the research advances that you mentioned, a lot of which are your contributions to projects like Centauril, along with Nvidia's tools (like YZX, etc.) have greatly accelerated progress in this area, especially in the area of sensors and cheaper sensor hardware, etc. . So I feel like it's a very exciting time to be working in this area. Jim Fan Yes, I agree. Sonya Huang I remember your early research was more in the area of virtual worlds. Can you talk about what got you interested in Minecraft and robots? Do you think they're related? What got you interested in virtual worlds? Jim Fan That's a good question. For me, my personal mission is to solve the problem of embodied intelligence, and embodied intelligence agents in virtual worlds are agents like those found in games and simulations.