Figure AI's Figure 02 humanoid robot represents a significant leap in industrial robotics, powered by a new AI system called Helix that enables the robot to manipulate objects it has never encountered before without any special programming. The 5-foot-6-inch tall robot, which weighs 70 kilograms and costs approximately $130,000 per unit for commercial lease, has already been tested at BMW's Spartanburg plant performing chassis assembly tasks alongside human workers.

What Makes Figure 02's AI System Different From Other Robots?

The Helix Vision-Language-Action (VLA) model is the breakthrough technology behind Figure 02's capabilities. Unlike traditional robotics systems that rely on pre-programmed movements or narrow task-specific training, Helix uses a dual-system architecture inspired by how humans think and act. This architecture combines high-level scene understanding with real-time motor control, allowing the robot to adapt to new situations on the fly.

Figure AI ended its partnership with OpenAI in 2025, stating that large language models were "getting smarter yet more commoditized." The company then developed Helix in-house using approximately 500 hours of diverse teleoperated behavior data, with automatic labeling systems generating natural language instruction pairs for each behavior segment.

The system operates through two distinct processing layers. System 2 handles high-level scene understanding and task planning at 7 to 9 cycles per second, while System 1 executes precise motor actions at 200 cycles per second. This separation allows Figure 02 to be both general-purpose and fast, resolving a fundamental tradeoff in robotics where most systems must choose between being broadly capable or being quick.

How Does Figure 02 Actually Perform in Real-World Tasks?

Figure 02's capabilities extend far beyond simple object picking. The robot can perform zero-shot manipulation, meaning it can pick up virtually any small household or industrial object, including thousands it has never encountered before, simply by following natural language prompts. In demonstrations, Figure 02 has successfully manipulated delicate glassware, crumpled clothing, scattered small items, and standard industrial fixtures without any pre-programming for each specific object.

The robot's five-fingered hands are equipped with 16 degrees of freedom across both hands, enabling precise, complex movements that approach human-level dexterity. These hands can carry payloads up to 25 kilograms, making them suitable for handling heavy industrial components, packages, and tools. When paired with Helix AI, the hands adapt grip strength and finger positioning in real-time at 200 cycles per second.

One particularly impressive capability is multi-robot collaboration. A single set of Helix neural network weights can run simultaneously on two robots, enabling them to collaborate on shared tasks such as sorting groceries together. All behaviors, including picking, placing, using drawers, and refrigerators, run on one unified set of weights with zero task-specific fine-tuning.

What Are the Key Technical Specifications That Matter?

• Physical Design: Standing 168 centimeters tall and weighing 70 kilograms, Figure 02 is deliberately sized to human proportions, allowing it to navigate environments designed for people without requiring infrastructure modifications.
• Battery and Runtime: A 2.25 kilowatt-hour lithium-ion battery integrated into the torso provides 5 hours of continuous operation with approximately 1.5 hours of rapid charging time, representing a 50 percent improvement over the previous Figure 01 model.
• Computing Power: The robot features dual NVIDIA RTX GPU modules, providing roughly three times the computing power of Figure 01, with all Helix inference running entirely on the robot's embedded GPUs without requiring cloud connection.
• Sensory Systems: Six RGB cameras, inertial measurement units, gyroscopes, force sensors, touchless human detection, microphones, and speakers enable comprehensive environmental awareness and interaction capabilities.
• Structural Durability: Custom electric motors are integrated directly with joint drivetrains at each of the 28 degrees of freedom, with internal cabling routed to minimize wear during repetitive industrial motions.

The matte black exterior with fully integrated cabling provides significantly better protection against wear and damage during industrial operations compared to the exposed chrome elements and visible wiring of the previous generation. The 70-kilogram weight provides a low center of gravity that enhances stability during load-carrying tasks.

How Does Figure 02 Compare to Competing Humanoid Robots?

Figure 02 demonstrates a significant capability gap compared to most competitors in the humanoid robotics space. While the Tesla Optimus Gen 2 has shown improving hand dexterity, it hasn't demonstrated the same breadth of zero-shot manipulation capabilities. The Unitree G1 focuses more on locomotion than manipulation precision, making Figure 02's emphasis on hand dexterity and object manipulation a distinguishing factor in the market.

The robot's ability to handle objects it has never seen before without pre-programming represents a fundamental shift in how industrial robots can be deployed. Rather than requiring extensive training and customization for each new task or object type, Figure 02 can adapt on the fly, potentially reducing deployment time and increasing flexibility in manufacturing environments.

What's Next for Figure AI's Robotics Platform?

Figure AI has already announced Figure 03, the next-generation successor to Figure 02, which is scheduled for October 2025. The Figure 03 brings tactile sensors, wider field-of-view cameras, and 9 percent less mass than Figure 02, though Figure 02 remains in active deployment at industrial facilities.

The successful testing at BMW's Spartanburg plant demonstrates that Figure 02 is moving beyond research and development into real-world industrial applications. This transition from prototype to production-ready system marks a significant milestone for the humanoid robotics industry, suggesting that the technology is approaching practical viability for manufacturing and assembly tasks that have traditionally required human workers.