Watch: Figure’s humanoid robots work for 200 hours, process 250k packages without failure

Figure AI has demonstrated a significant milestone in humanoid robotics: its Figure 02 robots operated continuously for 200 hours in a warehouse setting, processing over 250,000 packages without a single failure. This achievement marks a leap toward commercial viability for humanoid robots in logistics, showcasing reliability and endurance that rival traditional automation.

Humanoid robots have long been proposed for warehouse and manufacturing tasks, but previous demonstrations often involved short, controlled runs or frequent failures. Figure AI’s test addresses the critical challenge of sustained operation in real-world conditions. Competitors like Tesla’s Optimus and Boston Dynamics’ Atlas have shown impressive capabilities, but Figure’s focus on long-duration, failure-free operation sets a new benchmark for deployment readiness.

During the 200-hour trial, the Figure 02 robots performed end-to-end package handling: picking items from conveyor belts, placing them into totes, and stacking totes onto pallets. The robots used vision-based perception and whole-body manipulation to adapt to varying package sizes and orientations. Key specs include a payload capacity of 20 kg per arm, a walking speed of 1.2 m/s, and a battery life enabling 8-hour shifts with quick-swap packs. The robots’ bipedal locomotion allowed them to navigate narrow aisles and step over obstacles, while their dexterous hands enabled precise gripping without pre-programmed motions. Figure’s design prioritizes modularity: the torso houses computing and power systems, while limbs are independently replaceable for maintenance.

The 250,000 packages processed without failure indicates a mean time between failures (MTBF) exceeding 200 hours, a critical metric for industrial adoption. Figure achieved this through redundant sensing (cameras, force-torque sensors, and IMUs) and a control architecture that blends model-based planning with learned reflexes. The robots operated in a shared workspace with human workers, using collision avoidance and speed limiting for safety.

Figure AI plans to deploy these robots in partner warehouses by late 2024, targeting e-commerce and parcel sorting facilities. The demonstrated reliability suggests humanoids could soon handle repetitive, physically demanding tasks alongside humans, reducing injury risk and labor shortages. For manufacturers, this validates humanoid form factors for logistics, potentially displacing fixed automation in environments requiring flexibility. Research labs will note the importance of long-duration testing for embodied AI, as the robots’ neural networks improved over the 200 hours through online learning.

Source: Interesting Engineering