Figure 02 at BMW Spartanburg: 30,000+ Vehicles, 1,250 Runtime Hours Validate...

Executive Deployment Metrics: BMW Figure 02 Pilot and GXO Digit Operations

In November 2025, Figure AI published results from its 11-month Figure 02 deployment at BMW Group Plant Spartanburg in South Carolina. Two Figure 02 units operated 10-hour weekday shifts for the full period, logging over 1,250 operational hours while loading more than 90,000 sheet metal parts with >99% placement accuracy against an 84-second cycle time target. The robots directly supported production of more than 30,000 BMW X3 vehicles. BMW described the results as demonstrating measurable added value under real-world conditions and announced evaluation of Figure 03 for expanded use cases, including European pilots at the Leipzig plant with Hexagon Robotics' Aeon platform for high-voltage battery assembly.

Concurrently, Agility Robotics' Digit platform achieved the first formal commercial RaaS humanoid deployment at GXO Logistics' Spanx fulfillment facility in Flowery Branch, Georgia. By mid-2026, the fleet had moved over 100,000 totes in live operations under a multi-year agreement signed in 2024. Digit units operate alongside humans in unmodified warehouse environments, handling tote transfer from autonomous mobile robots to conveyors. Additional Agility deployments include Schaeffler automotive plants and new agreements with Mercado Libre and Toyota Motor Manufacturing Canada.

These deployments represent the first production-validated data points for humanoid unit economics. BMW's pilot delivered consistent output without facility modifications, while GXO's RaaS model shifts capital expenditure to operational expense, enabling rapid scaling. Market forecasts project humanoid robotics for industrial logistics growing from $2.0 billion in 2025 to $14.8 billion by 2034, with automotive and 3PL segments driving 25-27% CAGRs.

Fleet Control Orchestration and Commercial Scaling Plans

Figure AI's Series C round, exceeding $1 billion at a $39 billion post-money valuation in September 2025 (led by Parkway Venture Capital with NVIDIA, Brookfield, Intel Capital, and others), funds BotQ high-volume manufacturing and the Helix embodied intelligence platform. The company targets shipping 100,000 humanoids over four years, leveraging BMW data as primary proof-of-performance for additional OEM sales. BMW and Figure are evaluating multi-unit expansion at Spartanburg plus geographic rollout to European plants.

Agility Robotics closed a $400 million Series C in 2025 at approximately $2.1 billion valuation, supporting RoboFab factory capacity targeting 10,000 units annually. Nearly 100 Digit units are deployed with paying customers including Amazon and GXO. The RaaS model charges subscription fees (industry estimates $2,000–$5,000 per robot monthly or $8–$18 per operational hour), eliminating upfront capital barriers for 3PL operators. Tesla's Optimus program advances separately: Gen 3 production slated for Fremont in summer 2026 after converting Model S/X lines, with internal fleets exceeding 1,000 units and targets of $20,000–$30,000 per unit. Boston Dynamics unveiled the production electric Atlas at CES 2026, with 2026 fleets committed to Hyundai's RMAC and Google DeepMind; Hyundai plans tens of thousands of units by 2028 via a dedicated 30,000-unit-per-year factory.

Orchestration platforms like Agility Arc enable remote monitoring, MES/WMS integration, and fleet management across multiple sites. Figure's approach emphasizes end-to-end data loops from BMW operations to refine policies. Total 2025 humanoid startup funding exceeded $3.2 billion globally, a 140% YoY increase, concentrated in US players with automotive and logistics anchors.

Technical Breakdown: Architecture, Actuators, and Sensors

Architecture Figure AI's Helix is a dual-system Vision-Language-Action (VLA) model. System 2 (S2) uses an onboard internet-pretrained VLM backbone operating at 7–9 Hz for scene understanding and language comprehension. System 1 (S1) employs a fast reactive visuomotor transformer policy translating latent representations into continuous actions at 200 Hz. The architecture supports zero-shot generalization across objects and collaborative multi-robot tasks using end-to-end training on robot data. Agility Digit leverages NVIDIA Isaac Sim for whole-body control foundation models trained across billions of simulation steps, emphasizing stability and recovery from disturbances in warehouse settings. Tesla Optimus and Boston Dynamics Atlas emphasize simulation-to-real transfer with reinforcement learning components for locomotion and manipulation.

Actuators and Sensors Digit features 20 actuators (brushless DC motors with custom transmissions), 5 DoF per leg and 3 DoF per arm, LiDAR for 3D mapping, four Intel RealSense depth cameras, RGB cameras, MEMS IMU/gyroscope for balance, absolute/incremental encoders, and arm force sensors for compliant manipulation. Payload capacity reaches 25 kg with battery runtime up to four hours and autonomous docking. Figure 02/03 platforms incorporate harmonic drives and high-torque electric actuators optimized for precision assembly tasks, paired with multi-camera vision suites and torque sensing. Industry-wide, FOC (field-oriented control) torque sensors and high-reduction harmonic gearing dominate for backdrivability and precision in collaborative environments.

Limitations Current drawbacks include energy density constraints limiting continuous runtime to 4–8 hours before charging, requiring frequent docking that disrupts 24/7 operations. Teleoperation latency and policy brittleness persist in unstructured edge cases, with fragility in high-DoF hands during prolonged contact tasks. Unit costs remain $150,000–$250,000 for early production units, though targets project $30,000–$50,000 within 3–5 years via volume manufacturing. MTBF data is emerging but limited; pilots report high uptime in structured environments yet require ongoing remote support. Cost per operational hour under RaaS approximates $8–$