Figure 02 Showcases Helix-Powered Household Autonomy

Autonomy Milestone Unveiled

Figure AI has unveiled groundbreaking footage of its Figure 02 humanoid robot autonomously unloading a dishwasher in under four minutes, powered by the newly developed Helix vision-language-action model. This demonstration, shared directly from company labs, marks a pivotal step toward everyday household integration, blending seamless perception, decision-making, and dexterous manipulation. Unlike scripted demos, Figure 02 navigates cluttered spaces, identifies dishware, and executes precise grips without human oversight, signaling readiness for labor-scarce homes and light commercial settings. CEO Brett Adcock highlighted this as proof of scalable generalist AI in physical form.

Helix Model Transforms Interaction

Helix unifies vision, language understanding, and action into a single neural network, enabling Figure 02 to interpret natural commands like "clear the dishwasher" and respond with fluid, context-aware movements. Trained on synthetic data via NVIDIA Isaac Sim, the model handles novel objects by generalizing from millions of simulated scenarios, achieving fine-grained control over 35 degrees of freedom at 200Hz. This empowers conversational fluency through integrated microphones and speakers, fostering intuitive human-robot teamwork. Early tests at BMW factories already validate its shift from rigid programming to adaptive, real-time problem-solving.

Compute Leap Accelerates Gains

Dual NVIDIA RTX GPU modules triple the on-device AI inference speed over Figure 01, processing complex visual data from six RGB cameras for robust object recognition and spatial mapping. Vision-based SLAM navigation pairs with IMU, gyroscope, and force sensors to deliver stable bipedal gait at 1.2 m/s, even amid dynamic obstacles. Compliant actuators and touchless human detection ensure safe coexistence, while a custom Linux OS with ROS-style APIs streamlines developer integration. These advances condense years of iteration into months, redefining humanoid viability.

Household Tasks Redefined

Beyond dishwashers, Figure 02 tackles laundry sorting, material handling, and logistics in real factories, proving its versatility for repetitive chores and hazardous duties. In BMW trials, it performed autonomous pick-and-place amid production lines, hinting at broader deployment in automotive assembly and warehouses. OpenAI partnerships enhance verbal tasking, allowing users to delegate via casual speech, much like chatting with an advanced assistant. This human-centric focus promises to alleviate domestic burdens, from elderly care to inventory management, ushering robots into lived spaces.

Dexterity Enablers Detailed

Six RGB cameras and force sensors grant Figure 02 human-like dexterity for gentle dish handling and precise placement, mimicking nuanced finger adjustments during unloading. Bipedal mobility at 3 km/h, supported by gyroscope and IMU, facilitates natural navigation through kitchens, enabling extended interaction sessions without fatigue. Compliant actuators absorb collisions, while 25 kg payload capacity empowers lifting heavy trays safely. Vision-SLAM and touchless detection boost collaborative skills, turning raw specs into reliable aid for dynamic human environments.

Rivals Edge Examined