Figure AI Is Supercharging Humanoid Robots— Here&#039, s How It Works

Find AI suddenly revealed on Thursday the “major miracle” that caused the fluttery robotics startup to cut ties with one of its investors, OpenAI: A book dual-system AI architecture that enables robots to view natural language commands and manipulate objects they’ve never seen before without needing any specific pre-training or programming for each one.

Helix combines a high-level reasoning method with real-time engine control, unlike regular drones, which require extensive programming or rallies for each new job. Its two systems efficiently bridge the gap between action or engine power, which is a knowledge of what objects are, and conceptual understanding ( knowing what they are ).

This will enable robots to improve their abilities over time without having to upgrade their devices or train with fresh data. The company released a video that shows two Number computers putting aside foods, one robot delivering products to the other machine, causing them to be placed in drawers and refrigerators to show how it works.

Number claimed that no machine was aware of the products they were manipulating, but that they were able to determine which ones should go in a freezer and which ones should be kept clean.

” Helix can relate to any home item”, Adcock tweeted. ” Like a people, Helix understands talk, causes through issues, and can understand any object—all without needing education or code”.

How the secret runs

To achieve this generalization capability, the Sunnyvale, California-based startup also developed what it called a Vision-Language-Action (VLA ) model that unifies perception, language understanding, and learned control, which is what made its models capable of generalizing.

This concept, Number states, marks many firsts in robotics. It outputs continuous control of an entire humanoid upper body at 200Hz, including individual finger movements, wrist positions, torso orientation, and head direction. Additionally, it enables two robots to work together with objects they have never seen before.

Helix’s breakthrough is due to its dual-system architecture, which mirrors human cognition: a 7-billion-percentile” System 2″ vision-language model (VLM) that processes high-level understanding at 7-9 Hz (updating its status with slow, structural tasks or movements ) and an 80-million-percentile” System 1″ visuomotor policy that translate those instructions into precise physical movements at 200 Hz ( basically updating its status 200 times per second ) for quick thinking.

Unlike previous approaches, Helix uses a single set of neural network weights for all behaviors without task-specific fine-tuning. One system processes speech and visual data to facilitate complex decision-making, while the other transforms these instructions into precise motor movements for real-time responsiveness.

” We’ve been working on this project for over a year, aiming to solve general robotics”, Adcock tweeted. ” Coding your way out of this won’t work, we simply need a step-change in capabilities to scale to a billion-unit robot level”.

Helix claims that all of this opens the door to a new scaling law in robotics, one that doesn’t rely on coding but instead relies on a collective effort to increase models ‘ abilities without any prior instruction on particular tasks.

Helix was figure-trained for roughly 500 hours of teleoperated robot behavior before creating natural language instructions for each demonstration using an automatic labeling system. The entire system is made possible for commercial use by the robots ‘ embedded GPUs.

Figure AI reported that it has already struck deals with a significant U.S. client named BMW Manufacturing. The company believes these partnerships create” a path to 100, 000 robots over the next four years”, Adcock said.

The humanoid robotics company secured$ 675 million in Series B funding earlier this year, from investors including OpenAI, Microsoft, NVIDIA and Jeff Bezos, at a$ 2.6 billion valuation. It’s reportedly in talks to raise another$ 1.5 billion, which would value the company at$ 39.5 billion.