• Funds Blog
• August 1, 2025

Helix: Breakthrough in Robotic Intelligence

Advertisements

In a groundbreaking announcement last Thursday, Figure AI unveiled Helix, a pioneering artificial intelligence framework aimed at empowering robots to comprehend their environments and perform complex tasks akin to human capabilities, without the necessity for specialized training. This revolutionary step comes shortly after the company severed ties with one of its key investors, OpenAI, indicating a bold shift in its strategic direction.

Founded as a startup focused on humanoid robotics, Figure AI has made significant strides in the realm of AI-driven mechanical entities. The recent launch of Helix highlights a formidable technological advancement that has propelled the company toward independence from collaborations that it previously deemed essential. CEO Brett Adcock remarked that the novel intelligence demonstrated by Helix allows Figure AI to advance its objectives without the support of OpenAI, just months after both companies announced their partnership during a lucrative funding round that raised a staggering $675 million, pushing Figure AI's valuation to $2.6 billion.

Interestingly, the company’s transformation has not gone unnoticed by the investment community. The B-round funding earlier this year included backing from formidable entities such as OpenAI, Microsoft, Nvidia, and Jeff Bezos. With ambitions of securing an additional $1.5 billion, Figure AI is poised to inflate its market valuation to an astounding $39.5 billion if successful. Such figures underscore the intense interest surrounding the automation and robotics sector, which is rapidly evolving as technology continues to advance.

At the heart of Helix is a unique Vision-Language-Action (VLA) model, which integrates perception, linguistic comprehension, and learning control mechanisms, offering the ability to generalize across various applications. This holistic approach marks a departure from conventional robotics, where machines typically require extensive programming or demonstration learning to tackle new tasks. Instead, Helix combines sophisticated reasoning with real-time motion control, effectively bridging the gap between semantic understanding—what it means to identify objects—and motion control—determining how to interact with these objects.

What sets Helix apart is its dual-system architecture that mirrors human cognitive processes. This architecture consists of a “System 2” visual-language model (VLM) with 7 billion parameters, designed for high-level comprehension tasks, operating at a frequency of 7 to 9 Hz, akin to “thoughtful deliberation”. Meanwhile, the “System 1” visuomotor control policy, boasting 80 million parameters, functions at a staggering 200 Hz, translating instructions into precise physical actions, reminiscent of “instinctive reactions”. This intricate balance between cognitive processing and motor control allows Helix to operate in a manner that has never been seen in robotics before.

The implications of this technological feat are profound. Helix enables robots to become increasingly proficient over time without the need for constant software updates or retraining on new data. In showcasing this technology, Figure AI released a video demonstrating two Figure robots collaboratively organizing groceries—one handing items over while the other expertly stores them into a drawer and a refrigerator. Remarkably, these robots had never encountered the items they were handling but still managed to discern which belonged in the fridge and which should be stored in a dry location.

“Helix can generalize to any household object,” Adcock stated via social media, emphasizing the framework's versatility. “Like humans, Helix understands spoken language, reasons through problems, and can grasp any object—without the need for training or extra programming.” This capacity for generalization represents a significant leap forward in the field of robotics, exemplifying a shift towards more adaptable, intelligent machines.

In the broader context of robotics and automation, Helix is touted as a series of firsts for the industry. It achieves continuous control of a humanoid robot's upper body, including finger, wrist, torso, and head movements at 200 Hz. The system also allows two robots to collaborate on tasks involving items they've never encountered, showcasing a level of cooperative functionality that mimics human teamwork.

Differentiating itself from traditional methods, Helix avoids the dependency on task-specific fine-tuning. Instead, it employs a singular set of neural network weights for executing all behaviors. One component of the system interprets auditory and visual data to facilitate complex decision-making, while the other translates these directives into precise motor actions, enabling instantaneous responses to real-world stimuli.

Adcock expressed the intense research efforts that led to this innovation, stating, “We have been working on this for over a year, with the goal of fundamentally solving the general robotics problem. Programming alone cannot achieve this; we need a qualitative leap in robotic capabilities to scale to a billion robots.” This ambition underscores a growing belief within the industry that the future of robotics lies not only in improved algorithms but in intelligent systems that can learn and adapt without constant manual input.

Helix ushers in a new paradigm in robotics, encapsulating the concept of "scalability by design"—moving away from programming reliance to a model of extensive collective learning that enhances the system's capabilities without targeted training. This marks a significant evolutionary point in how robots may function within society, from domestic assistance to industrial applications.

To train Helix, Figure AI utilized an extensive dataset of 500 hours of remotely operated robotic behaviors, employing an automatic labeling process to generate natural language instructions for every demonstration. This innovative approach allows Helix to operate on an embedded GPU within each robot, positioning it for immediate deployment in commercial settings.

Moreover, Figure AI has disclosed partnerships with BMW’s manufacturing division, as well as a major undisclosed client in the United States. These collaborations are anticipated to pave the way for the deployment of 100,000 robots over the next four years, a substantial step that realigns the landscape of robotics in industrial production and everyday life. This trajectory not only signifies a remarkable technological advancement but also heralds a future where intelligent robots become integral players across various domains.