Figure AI's latest humanoid robot, the Figure 03, has moved beyond simply proving it can work for hours without stopping, instead demonstrating the speed and adaptability that real warehouse floors demand. In a demonstration shared by Salesforce CEO Marc Benioff, the Figure 03 handled a package-sorting task with notable fluidity, even when Benioff deliberately tossed packages back to the robot to test its reactive intelligence. The robot adjusted its grip and continued working without missing a beat, revealing a significant leap in both speed and autonomous decision-making compared to its predecessor. What Changed Between Figure 02 and Figure 03? The jump from Figure 02 to Figure 03 represents a fundamental shift in how Figure is approaching the humanoid robotics challenge. While the Figure 02 focused on proving long-term endurance through a 60-minute unedited video, the Figure 03 prioritizes the speed and responsiveness that warehouse operators actually need. The Figure 02 achieved an average throughput of approximately 4.05 seconds per package, while the Figure 03 operates at a noticeably faster cadence, suggesting optimization improvements in how the robot processes visual information and plans its movements. The technical architecture underlying this performance jump reveals how Figure is rethinking robot control. Rather than relying on traditional rigid motion planning, the Figure 03 uses an end-to-end neural network that reasons directly from camera pixels to compute torque commands for its motors. This approach allows the robot to adapt fluidly to unexpected situations, like when Benioff interrupted the sorting task. How Does Figure 03 Actually Control Its Movements? - Motor Precision: The Figure 03 utilizes more than 30 motors across its limbs and extremities, providing fine-grained control that allows for complex manipulation tasks. - Direct Torque Computation: Instead of following pre-programmed motion sequences, the AI computes torque directly to control these motors, enabling real-time adaptation to changing conditions. - Visual Reasoning: The robot operates fully autonomously by processing camera input through a neural network that learns to map pixels directly to motor commands, bypassing the need for explicit instructions. This architecture matters because it allows the Figure 03 to handle the unpredictability of a real warehouse environment. When Benioff grabbed packages the robot had already sorted and tossed them back, the Figure 03 immediately re-evaluated the scene and handled the returned items without requiring human intervention or a reset. This kind of reactive behavior is essential for automation systems that need to work alongside human workers. "It's really impressive," noted Salesforce CEO Marc Benioff as the robot adjusted its grip and continued the task despite the interference. Marc Benioff, CEO at Salesforce Why Speed Matters More Than You Might Think? The shift from endurance testing to speed optimization signals a maturation in how Figure is thinking about commercial deployment. Warehouse automation isn't primarily about proving a robot can work for 60 minutes straight; it's about moving packages through a facility faster than human workers while maintaining accuracy. The Figure 02 had already improved barcode scanning accuracy to 95%, a solid benchmark. The Figure 03 builds on that foundation by adding the throughput and real-time responsiveness that make the technology viable for actual logistics operations. Brett Adcock, Figure's CEO, clarified the technical underpinnings of the Figure 03's performance in his response to Benioff's demonstration. The robot's ability to operate "fully autonomously" while handling dynamic interference suggests that Figure has solved a critical problem in warehouse automation: how to build a system that doesn't break when real-world conditions deviate from the training scenarios. The practical implication is significant. Warehouse operators don't care whether a robot can work for an hour without stopping if it can't handle the chaos of a real logistics floor. The Figure 03 appears to address both concerns, combining the durability proven by the Figure 02 with the speed and adaptability that make commercial deployment feasible. As Figure continues to iterate on its hardware and software, the company is moving closer to the point where humanoid robots become a practical alternative to traditional automation systems in warehouse environments.
