Figure AI Unveils Index: A Bold Plan to Mine Real-World Data for Robots

Figure AI, the embodied intelligence startup backed by OpenAI, NVIDIA, and other Silicon Valley giants, has finally unveiled its core strategy: Index. This is not another impressive hardware iteration, but an ambitious data engineering endeavor. Through 'global physical sampling' in the real world, the company aims to build the most diverse robot training dataset ever created. In essence, Figure AI is compiling a 'survival guide' for robots, teaching them to truly understand and integrate into the physical world.

The underlying logic is straightforward: the internet, full of cat videos and encyclopedic knowledge, cannot teach robots how to wash dishes or carry heavy objects. To build truly general-purpose robots, physical interaction data that does not exist online is essential. Figure's solution is to 'create it manually' through a system called 'Index Collect' (see image below). This system records first-person videos and fine-grained action data of humans performing daily tasks, and maps them in real time to the digital twin of its Figure 01 robot.