Figure 02 BMW deployment — How the vanguard of humanoid robots performed on the factory floor

Summary — quick numbers and headline takeaways

Figure (Figure) reports that its second‑generation humanoid, Figure 02, was deployed at BMW Group (Bǎomǎ Qítuán 宝马集团)ʼs Spartanburg plant over a six‑month period and participated in the assembly process of roughly 30,000 BMW X3 vehicles.

The robots loaded more than 90,000 parts, logged over 1,250 operating hours and covered an estimated 1.2 million steps (about 200 miles / ≈322 km).

• Participation: Figure 02 units helped produce ~30,000 BMW X3s over a six‑month stretch.
• Parts handled: Over 90,000 parts were picked and placed by the robots.
• Runtime: More than 1,250 hours of cumulative run time was recorded.
• Mobility: Robots logged ~1.2 million steps, roughly 200 miles (≈322 km).

Deployment highlights — what Figure disclosed from Spartanburg

• Participation: Figure 02 units helped produce ~30,000 BMW X3s over a six‑month stretch.
• Parts handled: Over 90,000 parts were picked and placed by the robots.
• Runtime: More than 1,250 hours of cumulative run time was recorded.
• Mobility: Robots logged ~1.2 million steps, roughly 200 miles (≈322 km).

How the robots were used on the line — workflow and integration

The team described a pick‑and‑place workflow where a Figure 02 unit retrieves stamped sheet‑metal parts from racks or bins and places them into welding fixtures for an industrial six‑axis robot to weld.

After welding, the subassembly is passed back to the main production line, making the humanoid a literal upstream collaborator in an existing automated cell.

Performance metrics and targets (KPIs) — what success looked like

Figure set specific KPIs for the pick‑and‑place tasks to measure real manufacturing readiness.

• Cycle time: Complete the full task (including opening welding‑fixture door and loading) in ≤ 84 seconds.
• Positioning accuracy: Target >99% per shift for loading all three stamped parts correctly.
• Human intervention: Target = zero interventions per shift (no staff pause/reset events).

Operational calculations you can’t ignore — derived metrics from the deployment

From the raw numbers disclosed, you can compute practical factory rates without introducing new data.

• Average parts per hour: 90,000 parts ÷ 1,250 hours ≈ 72 parts/hour.
• Steps per hour: 1.2 million steps ÷ 1,250 hours ≈ 960 steps/hour.
• Distance per hour: 200 miles ÷ 1,250 hours ≈ 0.16 miles/hour (~260 meters/hour).

Those numbers give a baseline for throughput, wear‑and‑tear expectations, and how humanoid units might be billed or compared to existing automation cells.

Problems identified and engineering lessons — real failure modes matter

Deployment exposed weaknesses, with the forearm being the subsystem with the highest failure rate.

Contributing factors included compact packaging, requirements for three degrees of freedom and thermal constraints—making the forearm a particularly challenging subsystem for reliability and heat management.

This mirrors broader industry difficulties: dexterous hands and forearms are among the hardest parts of humanoid robotics.

Tesla (Tèsīlā 特斯拉)ʼs Optimus program previously reported issues such as motor cooling shortfalls and limited component lifetimes; Elon Musk has emphasized that designing wrists and forearms is especially difficult because many of the muscles that control hand function sit in the forearm, making those sections mechanically and thermally complex.

Figure 02 retirement and the transition to Figure 03 — iterate or die

With the introduction of Figure 03, the company initiated a formal retirement program for Figure 02 units and recalled deployed units back to headquarters.

Lessons from Figure 02 will directly inform manufacturing flow, component architecture and mechanical design of Figure 03.

Notably, Figure said it thoroughly re‑architected electronics in the Figure 03 forearm and wrist by removing separate distribution boards and reducing dynamic wiring runs to reduce complexity, improve reliability and simplify thermal management.

Industry context and outlook — where humanoid robotics sits today

The broader humanoid robotics industry has not converged on a single set of standards or form factors.

Market observers including Dongfang Securities (Dōngfāng Zhèngquàn 东方证券) note ongoing debates over choices like biped versus wheeled lower limbs, the level of hand dexterity required, whether linear joints are necessary, component selection, and robot size.

That lack of consensus increases SKU proliferation, dilutes capital and talent, and slows the emergence of embodied intelligence at scale.

From an investment perspective, market sentiment toward robot mass production has cooled and the robotics sector has pulled back slightly.

Analysts expect clearer mass‑production signals by the first half of 2026; they view a V3 rollout and small‑batch production in Q1 2026 as an important validation that technical and supply‑chain consensus is beginning to converge.

What this means practically for investors, founders, and engineers — tactical takeaways

• Real deployments matter: The Figure 02 field test produced real factory hours, real part counts and measurable KPIs — this is rare and valuable evidence for any robotics playbook.
• Expect iterative recalls: Admitting subsystem reliability problems and issuing recalls highlights that hardware iteration is the norm, not the exception.
• Design for maintainability: Forearms and wrists are hotbeds for failure — simplifying electronics and thermal paths is a clear priority.
• Integration over replacement: Figure 02 acted as a collaborator inside an existing automated cell, which is a lower‑risk route for humanoids vs. attempting full line replacement.
• Signal watching for 2026: Investors should monitor V3 rollouts and small‑batch production plans in early 2026 as potential inflection points for the industry.

Bottom line — a sober, optimistic read

The Figure 02 Spartanburg deployment is an important milestone in embodied automation because it generated concrete production metrics and real operating hours in a demanding environment.

At the same time, the public recall and redesign underline that humanoid robotics still requires frequent, hands‑on iteration between field deployment and engineering refinement.

For stakeholders — investors, founders, engineers and automation leads — the right frame is pragmatic: treat these trials as accelerated R&D programs that produce measurable lessons rather than overnight factory replacements.

Figure 02 BMW deployment remains a useful data point as the industry pushes toward Figure 03 and beyond.

References

• 机器人“进厂”先头部队绩效如何？Figure披露：已参与3万辆宝马汽车生产 – 东方财富
• 财联社 (Cailian Press) – Home
• Figure 02 / Figure 03 information and company site – Figure
• Brett Adcock profile (social posts about Figure deployments) – X