Guotai Haitong Securities released a research report stating that Figure's Helix 02 has achieved a key technological breakthrough, advancing humanoid robots from technical demonstrations to practical applications. A single neural control system has solved the industry bottleneck of integrating mobility and manipulation. Looking at the broader industry, numerous humanoid robotics companies are continuously pursuing technological innovation and exploring real-world applications, collectively driving the industrialization of humanoid robots forward. The year 2026 is anticipated to be a critical milestone for the commercialization of humanoid robots. The focus should be on suppliers of core robotic components. Guotai Haitong's primary views are as follows.
On January 27, the humanoid robotics company Figure officially launched its new generation of self-developed AI model, Helix 02. Compared to the previous Helix model, which primarily controlled upper-body movements, Helix 02 marks a critical leap in Figure's capability for full-body autonomous control of humanoid robots.
The Helix 02 model has passed validation in core scenarios, pushing humanoid robots further from the laboratory toward practical use. In the released video, the Figure 02 robot autonomously completed the task of loading and unloading a dishwasher—a sequence of 61 coordinated actions without human intervention—demonstrating potential for deployment in settings like home services and restaurant kitchens. The robot can autonomously perform actions such as walking, picking up and placing items, and stacking, and can even utilize its hips and feet for cooperative tasks when its hands are occupied. In the realm of fine manipulation, leveraging tactile sensing and visual feedback from its palm, it successfully accomplished tasks like unscrewing a bottle cap, extracting a single pill, precise liquid dispensing, and sorting in cluttered environments, overcoming the limitations of pure vision-based control. This breakthrough also opens up technical possibilities for subsequent high-precision scenarios such as medical assistance and precision manufacturing. The video demonstration also showcased the robot's ability to adapt to dynamic changes in a real environment, possessing error-correction capabilities, indicating potential for stable task execution in complex situations.
The key breakthrough of Helix 02 lies in its creation of a "three-level协同" integrated control architecture. The newly added SYSTEM 0, operating at a 1kHz frequency, governs full-body balance and coordination. Trained on over 1,000 hours of human motion data and reinforced through 200,000 parallel simulation environments, it uses a single neural network to replace millions of lines of hand-coded instructions, achieving human-like natural motion. SYSTEM 1 has been upgraded to a "full-sensory input to full-actuation output" visuomotor policy, integrating highly sensitive tactile sensors, multi-dimensional cameras, and full-body proprioception to build an end-to-end pathway from pixels to joint control. It balances millimeter-level precise manipulation with room-scale mobility, spanning a dynamic range of four orders of magnitude. SYSTEM 2's semantic reasoning capability has been enhanced, enabling it to parse long-horizon, complex instructions without needing to intervene in low-level motion planning. Overall, Helix 02 addresses the industry pain point of "mobility and manipulation constraining each other." Although the current achievements are still in early stages, they fully demonstrate the immense potential of full-body autonomous control.
Risk warnings include technological progress and industrialization advancing slower than expected, and intensifying industry competition.
Comments