Figure 02 and Helix AI: A Leap Toward Autonomous Humanoid Robotics

Figure 02 and Helix AI: A Leap Toward Autonomous Humanoid Robotics
In a groundbreaking demonstration of artificial intelligence and robotics, Figure AI has unveiled the Figure 02 humanoid robot, powered by its proprietary Helix Vision-Language-Action (VLA) model. This advanced system enables the robot to perform complex tasks autonomously, offering a glimpse into a future where humanoid robots could become integral to daily life. A recent video shared by Figure AI’s CEO, Brett Adcock, showcases the Figure 02 sorting and scanning packages in real time, flipping them to orient barcodes correctly and even flattening soft packages for scanning—tasks typically performed by human workers. This article explores the capabilities of Figure 02, the innovative Helix AI, and the broader implications for industries and households, as well as the ethical and societal challenges that lie ahead.
The Figure 02: A Humanoid Designed for Versatility
Founded in 2022, Figure AI has rapidly emerged as a leader in humanoid robotics, with a mission to create general-purpose robots capable of performing diverse tasks in industrial and domestic settings. The Figure 02, introduced in August 2024, represents a significant evolution from its predecessor, Figure 01. Standing at a human-like height, it features six RGB cameras, a 50% increased battery capacity (up to 10 hours of runtime), and hands with 16 degrees of freedom capable of lifting 25 kilograms. Its sleek design integrates cabling within limbs and a torso-embedded battery, optimizing both aesthetics and functionality. Powered by NVIDIA RTX GPU modules, Figure 02 boasts three times the computational power of Figure 01, enabling sophisticated AI processing directly on the robot.
The robot’s hardware is complemented by advanced conversational capabilities, initially developed in collaboration with OpenAI but now driven by Figure’s in-house Helix model. With microphones and speakers, Figure 02 can engage in natural dialogue, making it suitable for human-robot interactions in dynamic environments. Its ability to move at 1.2 meters per second and perform precise manipulations, such as grasping delicate objects or operating machinery, positions it as a versatile tool for both structured industrial settings and unpredictable home environments.
Helix AI: The Brain Behind the Brawn
At the core of Figure 02’s autonomy is Helix, a Vision-Language-Action model that integrates visual perception, natural language understanding, and precise motor control. Unlike traditional robotic systems that require extensive programming or task-specific training, Helix enables Figure 02 to generalize behaviors dynamically, responding to natural language prompts without prior demonstrations. This breakthrough, announced in February 2025, marks a significant departure from conventional robotics, which often struggle with adaptability in unstructured settings.
Helix operates through a dual-system architecture inspired by human cognition:
- System 2 (S2): A 7-billion-parameter Vision-Language Model (VLM) pretrained on internet-scale data, S2 handles high-level reasoning, scene understanding, and language processing at 7–9 Hz. It interprets abstract commands, such as “Pick up the desert item,” and identifies relevant objects, like a toy cactus, in cluttered environments.
- System 1 (S1): An 80-million-parameter visuomotor policy, S1 translates S2’s directives into real-time actions at 200 Hz, controlling the robot’s upper body, including torso, wrists, and individual fingers, with 35 degrees of freedom. This high-rate control ensures precise, human-like dexterity.
Trained on approximately 500 hours of teleoperated behaviors, Helix uses an auto-labeling VLM to generate natural language instructions from video data, reducing the need for extensive datasets. Its ability to run on low-power embedded GPUs makes it commercially viable, eliminating reliance on external computing resources. In demonstrations, Helix-powered Figure 02 robots have handled thousands of novel household objects—glassware, toys, tools, and clothing—without task-specific fine-tuning, achieving a transformative leap in robotic adaptability.
Real-World Applications: Logistics and Beyond
The video shared by Brett Adcock on X highlights Figure 02’s prowess in logistics, a critical application for Figure AI’s commercial strategy. In an unedited, hour-long demonstration, the robot sorts packages at a warehouse, processing each in approximately 4.05 seconds—a 20% improvement over previous performance—with a 95% success rate. Using its six RGB cameras and stereo vision, Figure 02 identifies and orients barcodes, adjusts for package variations (from rigid boxes to deformable bags), and self-corrects errors in real time. A feature dubbed “Sport Mode” accelerates movements by up to 50%, enhancing throughput while maintaining precision.
This capability stems from Helix’s advanced features:
- Implicit Stereo Vision: Enhanced 3D understanding enables precise depth-aware motion, crucial for handling dynamic objects on conveyor belts.
- Multi-Scale Visual Representation: S1 captures fine details and broader contextual cues, improving manipulation accuracy.
- Learned Visual Proprioception: Each robot calibrates itself, ensuring seamless policy transfer across multiple units without extensive retraining.
Figure AI has already secured partnerships with major companies, including BMW, where Figure 02 robots are being tested for tasks like inserting sheet metal parts in automotive manufacturing. Discussions with potential clients like UPS suggest further expansion into logistics and warehousing. The company’s second commercial customer, speculated to be Amazon or Walmart, underscores the growing demand for humanoid robots in structured environments. Figure aims to ship 100,000 robots over the next four years, leveraging its BotQ facility, which plans to produce 12,000 units annually.
Beyond industrial applications, Figure AI is accelerating its timeline for home deployment. CEO Brett Adcock announced that alpha testing of Figure 02 in residential settings will begin in late 2025, two years ahead of schedule, driven by Helix’s rapid advancements. In a demo, two Figure 02 robots collaborated to store groceries, responding to vague prompts like “Can you put these away?” They identified items, placed them in refrigerators and cabinets, and closed drawers without explicit instructions, showcasing their potential for household tasks.
Implications for the Future
The introduction of Figure 02 and Helix signals a paradigm shift in robotics, moving from task-specific machines to general-purpose humanoids capable of learning and adapting in real time. This has profound implications for industries and society:
- Industrial Transformation: In manufacturing and logistics, Figure 02 can address labor shortages and increase efficiency. Its ability to handle diverse packages and collaborate with other robots could streamline operations in warehouses and factories, reducing costs and improving throughput.
- Household Integration: In homes, Figure 02 could assist with chores, from tidying up to meal preparation, and support aging populations by enabling independent living. Its ability to navigate cluttered, unpredictable environments makes it a promising companion for domestic settings.
- Economic Impact: The rise of humanoid robots could stimulate job creation in robotics development and maintenance while raising concerns about job displacement in sectors like warehousing and retail. Figure AI emphasizes transparency to manage expectations and foster public acceptance.
However, challenges remain. Homes present unique obstacles—uneven lighting, varied furniture, and unpredictable human and pet interactions—that demand robust adaptability. Alpha testing will likely focus on simple tasks to refine Helix’s performance in real-world conditions. Pricing is another hurdle; Figure aims for a cost comparable to a modest vehicle, but achieving affordability without compromising functionality will be critical for