Helix 02: Full-Body Autonomous Humanoid Robot by Figure AI

Figure AI unveiled Helix 02 in January 2026, introducing the world’s most advanced full-body autonomous humanoid robot. This groundbreaking system marks a pivotal moment in robotics by seamlessly integrating walking, manipulation, and balance into a single neural network—achieving what researchers call true “loco-manipulation.”

Revolutionary Three-Tier Architecture

Helix 02 operates through an innovative System 0, System 1, System 2 architecture that mimics human cognitive processing at different speeds.

System 0 forms the foundation, running at 1,000 Hz to handle balance and coordination. This 10-million-parameter neural network learned from over 1,000 hours of human motion data, replacing 109,504 lines of hand-coded software with a single learned model. Trained across 200,000 parallel simulation environments, System 0 enables natural, human-like movement across the robot’s entire body.

System 1 operates at 200 Hz, connecting all sensors directly to all actuators. It processes input from head cameras, palm cameras, fingertip tactile sensors, and proprioception to control the complete robot—legs, arms, torso, and individual fingers. This unified “pixels-to-whole-body” approach allows the robot to reason about its entire state as one integrated system.

System 2 runs at 7-9 Hz, handling scene understanding and language processing. It interprets complex commands like “Walk to the dishwasher and open it” or “Carry the bowls to the counter,” translating human intent into actionable goals for the lower systems.

Longest Autonomous Task Ever Demonstrated

The most impressive demonstration of Helix 02’s capabilities is a continuous four-minute kitchen task—the longest horizon autonomous sequence ever completed by a humanoid robot. The robot walks to a dishwasher, unloads dishes, navigates across a full-sized kitchen, stacks items in cabinets, loads the dishwasher, and starts it—all without human intervention.

This achievement involves 61 distinct actions executed flawlessly, demonstrating several breakthroughs:

• Walking while carrying objects: The robot maintains stable grasps on delicate items through every step, coordinating forces across its entire body.
• Whole-body tool use: When hands are occupied, Helix 02 closes drawers with its hip and lifts doors with its foot, using the entire body creatively.
• Motor control across scales: The same neural network produces millimeter-precise finger movements and room-scale locomotion—a range spanning four orders of magnitude.

Advanced Dexterity with Tactile Sensing

Figure 03’s embedded tactile sensors detect forces as small as three grams—sensitive enough to feel a paperclip. Combined with palm cameras providing in-hand visual feedback, Helix 02 performs manipulation tasks previously impossible for robots:

Unscrewing bottle caps with precise torque control and tactile-regulated grip force, preventing crushing or slipping.

Extracting individual pills from medicine organizers, using palm-level vision when pills are hidden from head cameras.

Dispensing exact syringe volumes (5ml) through force-controlled actuation with multi-finger stabilization.

Picking metal components from clutter, handling overlapping objects that shift during interaction—tested on real parts from Figure’s manufacturing facility.

Learning from Human Motion Data

Unlike traditional robots requiring extensive programming for each task, Helix 02 learned natural coordination by studying over 1,000 hours of human movement. This approach produces fluid, intuitive motions that generalize to new situations without manual coding.

The system runs entirely on embedded GPUs, enabling real-time autonomous operation without cloud connectivity. This onboard processing ensures reliable performance with minimal latency—critical for responsive balance and manipulation control.

Real-World Applications

Helix 02’s capabilities open possibilities across multiple sectors:

Home assistance: Managing household chores from dishwashing to grocery organization, adapting to different room layouts naturally.

Healthcare: Assisting with medication management and patient care, leveraging precise force control for safe human interaction.

Manufacturing: Handling complex assembly tasks requiring both mobility and precision manipulation in facilities like Figure’s BMW partnership.

Logistics: Picking orders, restocking shelves, and organizing inventory with minimal infrastructure modification.

What Makes Helix 02 Different

Previous humanoid robots demonstrated impressive short-horizon behaviors like jumping or dancing, but most weren’t truly steerable—they replayed pre-planned motions with limited feedback. Helix 02 continuously perceives, decides, and acts in real-time, adjusting balance while reaching and recovering from unexpected situations autonomously.

Traditional systems separated walking and manipulation into distinct controllers stitched together awkwardly. Helix 02’s unified neural approach eliminates these artificial boundaries, enabling natural coordination between all body parts simultaneously.

Technical Specifications

• 35 degrees of freedom with human-like wrists and individual finger control
• Next-generation vision: doubled frame rates, 4x lower latency, 60% wider field of view
• Embedded palm cameras and 3-gram-sensitive tactile sensors in each fingertip
• Dual-GPU onboard processing for real-time autonomous operation
• Three-tier neural architecture (System 0/1/2) operating at 1kHz, 200Hz, and 7-9Hz
• Trained on 1,000+ hours of human motion data across 200,000+ simulation environments

Industry Impact

Helix 02’s January 2026 unveiling demonstrates that true full-body autonomous humanoid robots are technically feasible. The four-minute dishwasher task required solving robotics’ hardest problems: continuous loco-manipulation, long-horizon planning, robust error recovery, and safe object interaction.

Market analysts predict the humanoid robot industry could reach $38 billion by 2035. Figure AI’s approach—replacing hand-coded systems with learned neural networks—shows a path toward scalable, generalizable capabilities that could address labor shortages across industries.

The Path Forward

Figure AI continues scaling training data and refining architectures to expand task complexity and execution speed. The company has announced plans for continued development throughout 2026, with focus on faster execution, better generalization to diverse environments, and longer operational time.

The same system handling precise pill extraction also coordinates room-scale navigation while carrying dishes—versatility suggesting these robots could eventually handle the vast majority of physical tasks in homes and workplaces. As Figure deploys more robots in real-world settings, the gap between demonstration and practical everyday assistance continues to narrow.

Helix 02 represents more than incremental progress—it’s evidence that the science fiction vision of helpful household robots is approaching reality.

---