How Did the Vanguard of Factory-Deployable Humanoids Perform? Figure Says Its Robots Took Part in Producing 30,000 BMWs

factory-deployable humanoids

Key Points

Quick snapshot — Figure 02 field deployment at BMW (Baoma 宝马集团)

Figure (Figure) shared a six-month deployment update for its second-generation humanoid, Figure 02, after trials at the BMW Group (Baoma 宝马集团) Spartanburg (Sipatǎnbǎo 斯帕坦堡) plant.

The update includes a mix of operational metrics, failure analysis, and lessons feeding the next version, Figure 03.

Headline results from the deployment

• Participated in the production of 30,000 BMW X3 vehicles.
• Handled more than 90,000 individual parts (9万个零件).
• Accumulated over 1,250 operating hours.
• Estimated to have walked more than 1.2 million steps — roughly 200 miles (≈322 km).

What Figure 02 actually did on the line — the use case

The robots executed repetitive load-and-place tasks on BMW’s production flow.

The workflow was: pick sheet-metal parts from racks or bins, position them into a welding fixture, then hand that fixture to an industrial six-axis welding robot.

After welding, the assembled part continued down the main production line.

How Figure measured success — performance metrics for the pick-and-place task

Figure used clear, practical metrics to measure performance for the pick-and-place operation.

• Work cycle time — total time including opening the welding-fixture door and loading: target ≤ 84 seconds.
• Positioning accuracy — proportion of cycles in which all three sheet-metal parts were correctly loaded: target > 99% per shift.
• Human interventions — number of times staff must pause or reset the robot: target = zero per shift.

Failure points identified: the forearm and dexterous hand

Figure reported the robot’s forearm as the single most failure-prone subsystem during this deployment.

The company attributed failures to the forearm’s tightly packed mechanical and electrical layout, the need for multi-axis flexibility (three degrees of freedom), and thermal-management limits.

This mirrors a broader industry reality where dexterous hands and forearms are among the hardest engineering problems for humanoid robots.

Tesla (Tesla 特斯拉) and CEO Elon Musk (Masike 马斯克) have previously flagged similar issues for Optimus, including motor cooling shortfalls and limited part lifetimes.

Figure 02 retirement and lessons feeding Figure 03

With the public launch of Figure 03, Figure has initiated a formal retirement program for Figure 02 units.

Figure recalled related machines from BMW’s headquarters so field experience can inform manufacturing processes, component architecture, and mechanical redesigns for Figure 03.

Key changes called out for Figure 03 include a wholesale rework of the forearm and wrist electronics.

The redesign removes separate distribution boards and dynamic cabling in favor of a simplified architecture intended to reduce complexity, improve reliability, and make thermal management easier.

Industry context: consensus still forming around humanoid robot architecture

Analysts note the humanoid-robot industry has not converged on standard architectures or use-case definitions yet.

Dongfang Securities (Dongfang Zhengquan 东方证券) highlighted unresolved debates over:

• Mobility form factor — bipedal feet vs. wheeled bases.
• Design of dexterous end effectors (hands).
• Necessity and design of linear joints.
• Component selection and overall robot sizing.
• Managing a proliferation of SKUs, which can scatter capital and slow development of embodied intelligence.

From an investment perspective, Dongfang Securities says market confidence in near-term mass production has cooled and the robotics sector has seen a modest pullback.

They argue true mass production requires industry convergence on scenarios, models, and structures.

Dongfang Securities expects clearer mass-production signals in the first half of 2026, with a successful V3 rollout and small-batch production in Q1 2026 serving as important milestones.

What this means for investors, founders, and industrial teams

This deployment is a practical, early data point showing humanoids can be integrated into complex vehicle production workflows for repetitive load-and-place tasks.

The deployment also shines a spotlight on where engineering effort must go next:

• Reliability over novelty.
• Simplified mechanical/electrical architectures that ease thermal management.
• Focused use cases that play to robot strengths and reduce SKU fragmentation.

For investors, the takeaway is cautious optimism mixed with a clear signal to follow hardware roadmaps and field retirement/replacement plans closely.

For founders and product teams, the lesson is to prioritize subsystem robustness — especially in forearms and hands — and to design for maintainability in real factory environments.

For industrial adopters and plant managers, the data point suggests humanoids are already useful for routine, repetitive tasks, but long-term ROI will depend on lower intervention rates and predictable part lifetimes.

Practical checklist for teams evaluating humanoid pilots

• Define tight performance metrics up front — cycle time, accuracy, and human intervention thresholds.
• Track subsystem MTBF (mean time between failures) for modules like the forearm and wrist.
• Plan for field recalls and learn loops — treat pilot robots as iterative hardware releases.
• Focus on integration points with existing industrial robots (e.g., handoff to six-axis welders).
• Limit SKU creep — start with a single, repeatable use case like load-and-place.

Bottom line: cautious progress for factory-deployable humanoids

Figure’s update provides concrete proof that humanoids can be deployed in vehicle assembly lines for specific, repetitive tasks.

It also highlights the engineering hairballs — especially in forearm and wrist design — that will determine how fast the industry moves from pilots to mass production.

The route to commercially meaningful, large-scale factory deployment will likely run through simplified, robust subsystems and industry convergence on a narrower set of use cases and hardware conventions.

factory-deployable humanoids

References

• Robots’ factory vanguard performance: Figure discloses participation in 30,000 BMW cars production – Cailian Press
• Brett Adcock repost about Figure 02 field deployment – Brett Adcock
• Figure Labs official site – Figure