Figure 03 vs Apptronik Apollo 2: Which Is Further Ahead?

Figure 03 and Apptronik Apollo 2 were both put in the spotlight on June 30, 2026. Figure showed its latest robot handling a sequencing task at BMW. Apptronik unveiled Apollo 2 as the workhorse behind its expanded Robot Park and customer data-collection programs.

They look like close rivals, but the public evidence points to two different products. Figure 03 is a bipedal platform that Figure is already building in the hundreds and placing into home research and commercial work. Apollo 2 is a modular training and pilot platform, offered with legs or a wheeled base, whose data is meant to improve Apptronik's next commercial generation.

Verdict: Figure publishes the stronger manufacturing and product-readiness case today. Apollo 2 is the more flexible data-collection platform, but Apptronik's own announcement makes clear that its work feeds the coming Apollo 3 fleet. Neither robot has a public list price or open order page.

Figure 03 vs Apollo 2 at a glance

The table deliberately does not copy the familiar 2023 Apollo numbers into the Apollo 2 column. Apptronik has not confirmed that the new platform keeps the older robot's height, weight, payload, speed or battery runtime. Treating those generations as one spec sheet would make the comparison look more precise than the evidence allows.

The biggest difference is product purpose

Figure presents Figure 03 as its current general-purpose robot. It was redesigned for Helix, home environments and high-volume manufacturing, but Figure also says the same hand, sensor and charging changes support commercial work. Figure's June 2026 demonstration showed precise part handling, full-body movement and cart pulling during a sequencing task.

Apptronik describes Apollo 2 differently. The company calls it the workhorse behind Robot Park, where robots collect data across logistics, manufacturing, retail and other customer-led tasks. The platform is already present at partner and customer sites, but its stated job is to create the experience and training data that will power Apollo 3.

That does not make Apollo 2 a dead-end prototype. It makes the buying question narrower. Apollo 2 looks relevant for a company that wants to join a structured pilot or embodied-AI data program. Figure 03 looks closer to a single hardware generation that Figure intends to operate, improve and manufacture at scale.

Hardware: fixed biped versus modular mobility

Figure 03 is bipedal only. That gives it access to stairs, narrow passages and workstations designed around a standing person. Figure's BMW example also shows why legs can matter during a sequencing job: the robot steps, shifts its reach and repositions its whole body while handling parts and moving a cart.

Apollo 2 can use legs or a wheeled base. The biped is meant for spaces where human-like mobility matters. The wheeled version gives up stairs and some terrain flexibility in exchange for stability, simpler movement and easier alignment with existing industrial mobile-robot safety practices.

For a flat warehouse or factory, the wheeled Apollo 2 may be the more practical pilot platform. If the task genuinely needs a robot to move through human spaces, reach around fixtures and coordinate manipulation with foot placement, Figure 03 has published the clearer demonstration.

AI: Helix 02 versus Artemis and Gemini Robotics

Figure owns the full AI stack it discusses publicly. Helix 02 coordinates hands, arms, torso and feet through a pixels-to-actions model. At BMW, Figure says the system adjusted to parts and carts that were not presented in exactly the same position each time. Figure also uses its growing robot fleet to collect data and push new behaviours through its fleet-management and over-the-air update systems.

Apollo 2 uses Apptronik's Artemis layer to coordinate perception, planning, controls, safety, task execution and human interaction. Fleet Connect handles monitoring and orchestration. Apptronik's research partnership with Google DeepMind adds Gemini Robotics models, while Robot Park supplies real-world data through teleoperation, autonomous execution and simulation.

There is no honest universal winner here. Figure has shown a more vertically integrated autonomy story on its own current hardware. Apptronik has a broader modular data engine and a major external AI partner. Buyers should ask for task success rates, interventions, recovery behaviour and operating hours on their exact workflow, not choose between two polished demo reels.

Manufacturing and deployment evidence

Figure says it had delivered over 350 Figure 03 robots by April 29, 2026 and had demonstrated a one-robot-per-hour cycle time at BotQ. It also reported an end-of-line first-pass yield above 80%, more than 500 battery packs and more than 9,000 actuators. Those are company-reported figures, not an independent production audit, but they are unusually concrete for this market.

BMW says Figure 03 will work on complex sequencing applications in logistics at its Spartanburg plant. The public evidence describes a project and demonstration rather than customer-verified throughput data, but it independently connects the current robot generation to a named factory and defined task.

Apptronik says operational Apollo 2 fleets are active in Robot Park locations and at partner or customer sites worldwide. It names Google DeepMind, Mercedes-Benz and GXO, and says the robots work across customer-driven tasks. Google DeepMind separately confirms its Apptronik partnership and Gemini Robotics work on Apollo. What is missing is equally important: Apptronik has not published the number of Apollo 2 units, total operating hours, task throughput or customer-confirmed performance.

On public evidence alone, Figure is ahead on disclosed manufacturing scale. Apptronik is ahead on mobility options and the breadth of its described data-collection network.

Power and uptime

Figure publishes a five-hour runtime for Figure 03 and says the robot can recharge at 2 kW by standing on an inductive mat. That design reduces manual battery handling and suits jobs where the robot can take planned charging breaks. Figure also pairs the dock with wireless data offload.

Apptronik chose swappable batteries for Apollo 2, along with opportunity charging and tethered operation. The company says this setup is intended to support long operating windows with minimal interruption. It does not currently publish a per-pack runtime for Apollo 2, so a direct hours-versus-hours comparison would be guesswork.

The trade-off is operational. Figure's dock is simpler for an autonomous fleet if the duty cycle leaves enough charging time. Apollo 2's battery swaps can keep a robot working while another pack charges, but they add battery inventory, handling and maintenance procedures.

Price and availability

Neither company publishes a list price for these robots. Figure 03 has no public order page. Apptronik invites companies to discuss Apollo deployments, but Apollo 2 also has no public checkout, reseller price or standard delivery promise.

That makes both enterprise engagements rather than normal equipment purchases. Before treating either as available, ask for a written scope covering the robot configuration, software, support, spares, remote access, site integration, acceptance tests and responsibility for failed shifts. Our guide to humanoid robots you can actually buy separates open sales from pilots and development programs.

Do not compare a rumoured hardware price in isolation. Battery packs, charging infrastructure, guarding or site changes, integration work, operator training and field service can dominate the total cost of a humanoid robot.

Which robot should you choose?

Choose Figure 03 for a serious bipedal pilot

Figure 03 is the stronger shortlist candidate when the task requires human-space mobility, dexterous manipulation and a vendor that can point to a disclosed production ramp. The BMW sequencing example is more useful than a generic capability claim because it shows the robot combining locomotion, force and precise placement in one workflow.

Choose Apollo 2 for modular pilot and data work

Apollo 2 makes more sense when a wheeled base would simplify the job, or when the goal is to participate in a structured data-collection and embodied-AI program. Its Artemis, Fleet Connect and Gemini Robotics stack is built around learning across multiple sites and configurations.

Choose neither for an off-the-shelf purchase

If you need a published price, standard delivery date and proven customer throughput, neither robot clears that bar. Ask both vendors for the same acceptance test and compare results on your site. A successful pilot should end with measured task completion, intervention rate, uptime, safety events and total operating cost, not a promise that the next software release will close the gap.