Figure AI has unveiled Helix, a sophisticated artificial intelligence model that allows humanoid robots to execute real-world tasks using voice commands, recognize and interact with objects they have never encountered before, and collaborate on assignments with other robots.
Unlike conventional robotics approaches that require extensive pre-programming, Helix enables robots to adjust to dynamic environments on the fly.
The Helix AI system merges vision, language processing, and motion control into a single neural network, eliminating the need for separate AI modules. Helix operates entirely on embedded GPUs, allowing robots to function without reliance on cloud processing, making real-time response times significantly faster.
The introduction of Helix follows Figure AI’s decision earlier this month to move away from OpenAI’s models in favor of developing a proprietary AI system optimized specifically for humanoid robotics. While OpenAI’s language models are well suited for general-purpose AI applications, they were not designed for the rapid, real-time decision-making required for robotic movement and object interaction.
Today, I made the decision to leave our Collaboration Agreement with OpenAI
— Brett Adcock (@adcock_brett) February 4, 2025
Figure made a major breakthrough on fully end-to-end robot AI, built entirely in-house
We’re excited to show you in the next 30 days something no one has ever seen on a humanoid
First Real-World Deployment: Figure AI’s Robots Are Already in Use
In December 2024, Figure AI transitioned from research to real-world deployment by delivering its first humanoid robots to a commercial client’s site. This milestone marked an important step toward proving the practicality of humanoid robots beyond controlled lab environments.
The integration of Helix into these deployed robots is expected to enhance their ability to take on a variety of tasks without requiring extensive retraining. Figure AI’s approach focuses on real-world commercial application, testing how humanoid robots can operate alongside human workers in industrial and service environments.
How Helix Improves Robot Adaptability
One of the biggest challenges in humanoid robotics has been enabling robots to interact with objects and execute tasks they were not explicitly trained for. Helix solves this problem through a Vision-Language-Action (VLA) model that enables robots to analyze their surroundings, interpret spoken instructions, and determine the best approach to executing a task.
Figure AI highlights Helix’s capabilities by explaining, “Helix displays strong object generalization, being able to pick up thousands of novel household items with varying shapes, sizes, colors, and material properties never encountered before in training, simply by asking in natural language.”
The AI system also supports multi-robot collaboration, enabling two or more humanoid robots to work together seamlessly. In testing, Helix-powered robots successfully coordinated to move and sort objects between them—an ability that significantly expands potential real-world applications.
By running exclusively on embedded GPUs, Helix ensures that all computations are processed locally, reducing latency and allowing robots to function autonomously without constant internet connectivity.
Unlike chat-based AI systems that process information sequentially, humanoid robots must continuously interpret sensory input, plan movements, and execute tasks with split-second adjustments. OpenAI’s models, which excel at text generation, lacked the precision and responsiveness necessary for Figure AI’s vision of humanoid automation. Developing an in-house AI allowed the company to optimize its systems specifically for robotic reasoning, spatial awareness, and adaptive control.
Figure AI’s Competitive Positioning in the Robotics Industry
Several major players are actively developing humanoid robots, but their approaches and priorities vary. Tesla’s Optimus is primarily focused on factory automation, with early prototypes performing structured assembly-line tasks under highly controlled conditions. Agility Robotics’ Digit has been designed for warehouse operations, particularly for handling logistics and package movement.
Boston Dynamics, which has long been a leader in humanoid mobility with its Atlas robot, continues to focus on refining movement and agility rather than full commercial deployment. While each of these companies is advancing different aspects of humanoid robotics, Figure AI is uniquely positioned by focusing on general-purpose humanoids that can work in multiple settings, including service industries, commercial operations, and potentially homes in the future.
Challenges That Still Need to Be Solved
Despite the advancements made with Helix, humanoid robots are still far from being fully autonomous assistants capable of handling any scenario without human intervention. One of the biggest remaining hurdles is fine motor control—while Helix allows robots to manipulate unfamiliar objects, tasks requiring delicate precision, such as folding clothes or handling fragile materials, still pose a challenge.
Another limitation is navigating unstructured environments. While Helix enables better object recognition and interaction, real-world spaces can be unpredictable, with obstacles, lighting changes, and human movement all affecting how a robot must adjust its actions. Further AI improvements will be necessary to ensure humanoid robots can reliably operate in dynamic, unscripted settings.
What Helix Means for the Future of Robotics
With Helix, Figure AI is making a strategic bet on humanoid robots becoming a viable part of the workforce in commercial and service settings. Unlike the rigid, pre-programmed robots traditionally used in manufacturing, Helix-powered robots are designed to learn in real time, collaborate, and handle varied tasks without needing extensive retraining.
Figure AI’s real-world trials will be crucial in determining whether humanoid robots can transition from experimental prototypes to widely deployed commercial solutions. The ability to function without cloud dependency and the introduction of multi-robot collaboration suggest that humanoid automation is getting closer to practical deployment at scale.
