Figure 03

Figure AI’s announcement, published June 30 2026, describes Figure 03 arriving in Hall 52 of BMW Group Plant Spartanburg to perform a “sequencing use case”: selecting and sorting individual automotive parts into delivery trolleys ahead of assembly.

The role combines precise picking and placement of thin-walled individual parts with forceful manipulation, including pulling loaded carts on caster wheels. Figure’s announcement frames this as more complex than the sheet-metal pick-and-place work performed by the prior generation.

Figure and BMW’s commercial agreement dates to January 2024. Figure 02 was credited with contributing to assembly of 30,000 vehicles at the Spartanburg plant in 2025. Coverage describes a contract scope with phased expansion to additional Spartanburg workstations through 2026-2027, plus separate pilot-stage arrangements at BMW’s Munich, Regensburg, and Leipzig facilities in Germany.

Figure’s announcement does not disclose unit counts, throughput rate, or task success rate for Figure 03 specifically at this task, and does not state whether any portion of the sequencing workflow uses teleoperation.

Figure AI’s BotQ manufacturing facility produced its 1,000th Figure 03 humanoid robot on July 23, 2026, according to eWeek. The company reports production throughput has increased from one unit per day to roughly one unit per hour, a roughly 24x increase achieved in under 120 days, a ramp first reported by TheAIInsider and RoboticsAndAutomationNews in May 2026.

Figure AI states end-of-line yield, the share of assembled robots passing final inspection, at over 80%. Battery production is reported at a 99.3% first-pass yield. Neither figure has been independently audited; both originate with company disclosure rather than third-party inspection.

BotQ is reported to be scaling toward an annual production capacity of 12,000 units. This is a stated target rather than a confirmed run-rate.

This entry documents Figure AI’s manufacturing output and factory throughput. It does not reflect any change to Figure 03’s software, hardware, or field-tested capability, and the robot’s Level II autonomy classification is unaffected.

Figure AI announced BotQ as a contract manufacturing partner for Figure 03. Manufacturer statements described BotQ as enabling production scale beyond internal prototype quantities. Specific unit volume targets, per-unit cost implications, and delivery timelines were not disclosed in published materials.

For development-phase humanoid platforms, in-house manufacturing represents a transition from hand-built prototype runs to repeatable production processes with direct quality control. Figure AI’s announcement indicated vertical integration as a deliberate step in the Figure 03 commercialization roadmap.

No hardware revision to Figure 03 specifications was announced in conjunction with the BotQ partnership. The manufacturing agreement applies to the existing Figure 03 design. Changes to tolerances or component sourcing during production transition were not disclosed.

Helix is a vision-language-action (VLA) neural network designed for end-to-end control on Figure 03. The model processes visual input and language prompts to generate motor commands without intermediate planning layers. Manufacturer documentation does not specify architecture details, training dataset composition, or model size.

Software-only update. No hardware revision required for Figure 03 units. Figure indicates the model runs on onboard compute with latency targets suitable for real-time control. Specific compute requirements and power draw not published.

Helix is trained on multi-task data to enable cross-task inference. Manufacturer claims demonstrate execution of manipulation tasks from visual input without task-specific fine-tuning. Performance on out-of-distribution tasks not disclosed. Few-shot adaptation capability mentioned but not benchmarked.

Figure 03 introduced revised torso, arm, and hand geometry compared to Figure 02. Manufacturer documentation cited improved range of motion and load capacity in the upper extremities. Specific joint torque ratings, degrees of freedom, and materials were not published at announcement.

Figure 03 integrates updated visual perception hardware positioned for Helix inference. Camera configuration and sensor specifications not disclosed. Onboard compute described as capable of running end-to-end neural network control in real time; hardware details not published.

Figure 03 was announced with Helix as the primary control architecture rather than as a retrofit. This represents a design-level integration distinguishing Figure 03 from prior-generation platforms that received Helix as a software update.

Manufacturer specifications list battery capacity at 2,300 Wh with approximately 300-minute runtime. Charging architecture not published at announcement.