BMW Group has expanded its testing of humanoid robotics by deploying Figure 03, a third-generation general-purpose machine developed by California-based Figure AI, at its vehicle assembly plant in Spartanburg, South Carolina. The deployment targets a specific sequencing use case within production logistics.
“Having already successfully completed a pilot with Figure 02 in our bodyshop, we are now looking forward to deploying Figure 03 for a sequencing use case in logistics,” stated Ulrich Wieland, vice-president of production control and logistics at BMW Manufacturing.
Technical Specifications: Figure 03 Mechanical and Power Profile
The third-generation humanoid platform introduces substantial hardware redesigns optimized for safety, mass production, and extended operation on the factory floor.
| Specification | Metric Value |
| Height | 1.73 metres (5 feet 8 inches) |
| Weight | 61 kilograms (9% mass reduction vs. predecessor) |
| Payload Capacity | 20 kilograms |
| Maximum Speed | 1.2 metres per second |
| Continuous Operating Time | 5 hours |
| Onboard Power Unit | 2.3-kilowatt-hour battery in the torso |
| Charging Mechanism | Wireless inductive coils built into the feet |
Kinematics, Sensory Feedback, and AI Architecture
The logistics assignment requires the robot to handle unsorted components arriving loose inside large containers. The machine must identify parts that have rotated or shifted, extract them, and place them into designated slots on a sequencing trolley before automated transport systems deliver the load to the assembly line.
[Unsorted Component Bin] ──> [Figure 03 Vision/Tactile Selection] ──> [Sequencing Trolley] ──> [Automated Tugger to Line]
To execute this, the robot utilizes a dual-layer vision system and highly precise tactile feedback networks:
- Vision Suite: Integrated head-mounted cameras map the broader workspace, while embedded palm cameras provide redundant, close-range visual data when the arms or containers block the primary sensors.
- Tactile Feedback: Proprietary fingertip sensors detect pressure as light as 3 grams, allowing the AI to distinguish between a stable grip and a slipping part. Softer, high-friction materials covering the fingertips maximize surface contact across varied component shapes.
- Full-Body Control: The machine dynamically adjusts its torso position and footing while executing complex extensions, reaches, or pulling tasks.
Operational logic is driven by the proprietary Helix 02 artificial intelligence framework. The system synchronizes visual, tactile, and spatial positioning data to orchestrate full-body motion. The software architecture splits computing cycles into three functional layers: an upper layer to interpret spoken commands and map task progression, a middle layer updating full-body movement trajectories 200 times per second, and a low-level control layer managing balance and motor actuators 1,000 times per second.
Hardware Modifications and Production Iterations
Figure AI utilized operational data from the previous Figure 02 pilot to eliminate high-wear mechanical vulnerabilities. The main structural changes focus on the forearm assembly, which had been the primary source of component fatigue:
Forearm Redesign: Engineers removed the independent distribution circuit board and moving internal cables that connected the wrist motors to the primary central computer. In Figure 03, the wrist motor controllers connect directly to the main processor, reducing internal part count and boosting component reliability.
The updated actuators operate at twice the speed of the prior generation while delivering greater torque density. For safer human-robot collaboration, the exterior features washable textile wraps and multi-density foam layers positioned around hard structural surfaces and mechanical pinch points. To facilitate large-scale manufacturing, Figure AI replaced individually machined structural components with stamped, cast, and injection-molded parts.
Predecessor Benchmarks: The Figure 02 Bodyshop Pilot
The logistics trial follows a rigorous 1,250-hour evaluation of the Figure 02 model within the Spartanburg factory’s welding and assembly division.
During that deployment, the automated system successfully moved more than 90,000 sheet-metal panels into fixed welding fixtures, contributing directly to the production of over 30,000 BMW X3 vehicles. The robot had to execute its task within a rigid 84-second factory cycle time, allocating 37 seconds to place three separate sheet-metal stampings. Certain tasks required a precise 5-millimeter tolerance completed within a 2-second window. Figure AI established a shift-level performance baseline of 99% successful placement without requiring manual intervention or human resets.
The companies have not publicly disclosed the exact number of active units deployed for the new logistics sequencing trial, the planned duration of the program, parts-per-hour output rates, purchasing costs, or hourly operating expenses.
