Figure 04 Design Lock Advances Custom Actuators

Brett Adcock's October All-Hands Announcement

Brett Adcock shared the Figure 04 update in a public recording of an internal all-hands meeting held around October 22, 2026. The presentation covered hardware milestones alongside data scaling and compute plans. Adcock positioned the new generation as the company's largest hardware leap yet, likening it to the original iPhone moment for humanoids. This comes after Figure built its 1,000th Figure 03 unit over the summer and achieved more than 90 percent cost reduction from Figure 02 to Figure 03.

The timing aligns with ongoing competition in humanoid deployments. Rivals such as Tesla Optimus and Unitree models continue production ramp discussions. Figure's approach emphasizes vertical integration in key subsystems. Observers note that design lock typically precedes prototype builds and early reliability testing.

In-House Actuator Development and Characterization

Figure is bringing actuator development in-house for Figure 04, with characterization of every unique actuator targeted for completion by the end of the week following the meeting. This shift moves away from reliance on external suppliers for core motion components. In-house work allows tighter integration with the robot's control software and potential improvements in thermal management and response times.

Actuator reliability directly affects mean time between failures in fleet operations. Custom designs could reduce downtime if characterization reveals and mitigates wear points early. However, internal production introduces new supply chain dependencies and requires specialized repair expertise at customer sites. Service economics will hinge on whether Figure maintains spare actuator inventories or offers rapid swap programs similar to those used in automotive manufacturing.

The decision mirrors strategies seen in other high-performance robotics programs where proprietary actuators improved efficiency but raised long-term maintenance costs. Figure has not disclosed actuator count, torque specifications, or expected lifespan metrics for the new units. Production planners will watch for these details ahead of any 2027 volume ramp.

Precision Hands with Millimeter-Level Tracking

The Figure 04 hand represents a fourth iteration focused on fingertip and contact surface tracking at millimeter precision against human references. Adcock described the hands as engineering masterpieces after months of refinement. This level of dexterity targets tasks requiring fine manipulation, such as handling small objects or delicate assembly work.

Dexterity gains matter for expanding addressable tasks beyond basic logistics. Improved contact sensing could lower failure rates in unstructured environments like warehouses or homes. Yet the added complexity may increase calibration time during initial setup and raise the skill level needed for field repairs.

Compared with prior generations, these hands aim for closer human parity without published benchmark scores such as those from standardized dexterity tests. Fleet operators evaluating total cost of ownership will factor in hand durability under repetitive cycles. Early data from Figure 03 deployments at sites like BMW could inform expectations for the upgraded version.

Index Dataset Scaling and Helix Model Investments

Figure's Index human-video program reached 37.5 million uploads across 112 countries by the time of the all-hands, up from roughly 16 million reported in August 2026. The company spent an estimated $8.1 million on data processing in a single week. App downloads reached 1.26 million, with 210,000 contributors uploading at least one video in the prior week.

This volume supports pre-training for the Helix family of models. Broader data diversity aims to improve generalization across unseen tasks and environments. The approach contrasts with narrower task-specific training used by some competitors.

Adcock outlined multibillion-dollar compute commitments, including an initial $3.5 billion Nscale arrangement with plans to scale toward $6 billion. The company holds more than $2 billion in cash. These investments target Helix 3.0 development following earlier versions.

Data acquisition costs at this scale raise questions about unit economics once robots reach commercial volumes. Operators will compare training expenses against expected productivity gains in deployed fleets. Figure has not released details on how Index data translates into measurable improvements in robot uptime or task success rates.

Production Ramp and Fleet Economics Outlook

Figure 04 component testing follows the design lock milestone. Earlier Figure 03 production reached one unit per hour at the BotQ facility. The new generation's in-house elements could support higher throughput if manufacturing yields stabilize quickly.

Serviceability implications include potential reductions in external part lead times for actuators and hands. Conversely, proprietary components may limit third-party repair options. MTBF data from Figure 03 treadmill tests, including one unit exceeding 348 miles, provide a baseline for reliability expectations.

Investors and customers will monitor whether the iPhone-level leap materializes in measurable metrics such as payload, runtime, or cost per unit. Without public pricing or delivery timelines, the announcement primarily signals engineering progress rather than immediate commercial availability.

Competitive Context and Next Steps

The update arrives amid active humanoid programs at Tesla, Boston Dynamics, Agility Robotics, and Unitree. Each pursues different balances of in-house versus off-the-shelf hardware. Figure's vertical integration on actuators and hands differentiates its path.

Helix model scaling through Index data could accelerate capability gains if compute investments deliver. Production economics will ultimately determine whether early pilots expand into large fleets. The October all-hands recording offers rare internal