Figure 02

Sheet metal handling at BMW Spartanburg

Figure 02 has been observed at BMW's Spartanburg, South Carolina plant placing sheet metal parts into fixtures on the body shop line. The deployment runs under a January 2024 Letter of Intent between Figure AI and BMW Manufacturing. Figure released video footage in August 2024 showing autonomous placement; BMW has not disclosed cycle times, defect rates, or how many shifts the robot covers.

End-to-end production with Helix VLA

Since February 2025, Figure 02 units run on Helix — Figure AI's in-house Vision-Language-Action model. Helix replaced the earlier OpenAI speech integration and runs entirely on the dual onboard NVIDIA GPUs without cloud dependency. Figure has shown demos of two robots sharing context to fold laundry and unload groceries, but the only confirmed production workload is the BMW body-shop task.

Logistics and warehouse pick-and-place

Figure has demonstrated tote handling and parcel manipulation in marketing footage. The 25 kg rated payload covers most parcel and tote weights up to typical e-commerce limits. Figure has not announced a named logistics customer comparable to the BMW deal. Apptronik (Apollo) and Agility (Digit) have publicly disclosed warehouse customers at GXO and Amazon respectively; Figure has not.

Bimanual tool use

Figure 02 has 16 degrees of freedom per hand, up from 6 on Figure 01. The August 2024 reveal showed bimanual tasks: passing objects between hands, holding tools steady while operating them. Throughput numbers have not been published. The hand revision is one of the most visible spec changes between generations.

Internal Helix data collection

Figure has positioned the BMW deployment as both productive work and a data-collection program. Each shift produces teleoperation and autonomy logs that feed Helix training. The mix of supervised teleoperation versus autonomous time on the BMW line has not been published by either Figure or BMW.

Figure 01 — Unveiled October 2023

First Figure walking prototype. Six-DoF hands, single set of cameras, dependent on remote OpenAI speech-to-speech for natural language interaction. Used for early Helix training data collection. Figure transitioned its public demo footage to Figure 02 in late 2024 and has not shipped Figure 01 to external customers.

Figure 02 — August 6, 2024

Second-generation platform with redesigned actuator stack, 16-DoF hands, six RGB cameras, dual onboard NVIDIA GPU modules, and a 2,250 Wh battery rated for 300 minutes of operation. Weight stayed at 70 kg; height at 168 cm. Initial software still routed conversational tasks through OpenAI.

Helix VLA replaces OpenAI integration — February 2025

Figure announced Helix, its own Vision-Language-Action model trained on internal teleoperation data. Running entirely on the dual onboard GPUs, Helix ended Figure's runtime dependency on OpenAI's API. This is a software upgrade for already-deployed Figure 02 units, not a new hardware revision.

Figure 03 — Announced October 2025, preorder

Smaller form factor aimed at home use, with a new wireless charging dock and revised hand design. Figure 03 is not yet shipping at the time of writing and detailed specs have not been published. Figure 02 remains the active enterprise platform. Production of both generations is planned at Figure's BotQ facility, with a stated four-year target of 100,000 units.

BMW Manufacturing, Spartanburg, South Carolina

The only publicly confirmed paying customer of any Figure platform. Tasks documented in Figure's own footage: sheet metal placement in body-shop fixtures. BMW's January 18, 2024 Letter of Intent did not commit either party to volume or duration. As of mid-2025 Figure described the deployment as still in early phase.

Source: Figure AI press release, January 18, 2024; deployment video, August 2024

Figure AI headquarters, Sunnyvale, California

Internal development and Helix training fleet. Figure has shown two-robot collaborative demos (laundry folding, grocery unloading) recorded at headquarters. The size of the internal fleet has not been published.

Source: Figure AI public videos, 2024-2025

Additional enterprise customers

Figure has not disclosed any other named customer. Brett Adcock (CEO) has referenced additional commercial agreements in podcast appearances and on X, but no second customer logo has appeared in Figure's official materials.

Single confirmed paying customer

BMW Spartanburg remains the only named commercial deployment. Apptronik (Apollo) at GXO, Agility (Digit) at Amazon, and Sanctuary (Phoenix) at Magna together cover several more named enterprise sites. Figure's deployment intensity, by public disclosure, is narrower than its $2.6B Series B valuation implies.

No ROS support and no published SDK

Figure 02 runs Figure's proprietary stack with Helix on top. There is no ROS or ROS 2 integration and no third-party developer program. Academic labs and integrators wanting open access default to Unitree H1, Agility Digit, or Boston Dynamics Atlas Research.

Vision-only sensing

Six RGB cameras, no LiDAR, no structured-light depth. This matches the Tesla Optimus approach and inherits the same limitations: known degradation in low-contrast and highly reflective environments. Figure has not published failure-mode analysis under factory lighting conditions.

Battery runtime is a manufacturer figure

The 300-minute runtime is a Figure-provided number. Duty cycle, payload conditions, and hand-actuation load are not specified. Real continuous-work runtime under sheet-metal handling on the BMW line has not been published.

Closed performance metrics

Figure has not published mean time between failures, joint wear data, fleet uptime, or shifts completed without operator intervention. Without these numbers, total cost of ownership cannot be compared against industrial robot disclosures from FANUC, ABB, or even Boston Dynamics' Spot reporting.

vs. Tesla Optimus Gen 2

Tesla and Figure share the vision-only sensing approach and the absence of ROS support, but the deployment posture differs. Figure has a named external customer (BMW); Tesla's roughly 300 Gen 2 units remain an internal fleet inside Fremont and Giga Texas. Tesla rates Optimus Gen 2 at 20 kg payload against Figure 02's 25 kg, and quotes a longer 480-minute battery runtime against Figure's 300. Tesla's stated long-term consumer price target is around $20,000; Figure 02 lists in the $130,000–$150,000 enterprise range. The two companies are pursuing the same end-state from different starting points: Figure ships to enterprises first, Tesla iterates inside its own factories.

vs. Boston Dynamics Atlas (Electric)

Atlas Electric replaced the hydraulic Atlas in April 2024 and runs on NVIDIA Jetson Thor with ROS 2 support — the opposite stack philosophy from Figure. Atlas rates 50 kg payload (double Figure 02), reflecting Boston Dynamics' decades of dynamic-manipulation research. Atlas Electric is deployed at Hyundai Motor Group, comparable in profile to Figure's BMW deal. The enterprise price gap is significant: Atlas Electric is estimated around $420,000 versus Figure 02's roughly $140,000 midpoint. Buyers choose Atlas for the gait library and dynamic capability; Figure for cost and the Helix VLA stack.

vs. Apptronik Apollo

Apollo and Figure 02 are the two US humanoids closest in spec: identical 25 kg payload, similar 173 cm / 72 kg form factor, and both targeting manufacturing and logistics. Apollo runs on Google DeepMind Gemini integration rather than an in-house VLA model. Apptronik has disclosed deployments at Mercedes-Benz and GXO Logistics — a broader named-customer footprint than Figure's single BMW site. Apptronik targets a sub-$50,000 enterprise price at scale, well below Figure's current $130,000+ range. Funding is the reverse: Figure raised $675M Series B in February 2024 at $2.6B valuation; Apptronik is smaller-funded but further along on confirmed customer logos.