Apptronik Opens Robot Park to Train Apollo 2
Real-world, task-level data remains the main bottleneck for training production-grade physical AI, so how a humanoid-robot maker chooses to generate that data is as important as the robot itself. Apptronik opened Robot Park, a nearly 90,000-square-foot facility in Austin, Texas, to collect task data from its humanoid fleet, according to a company press release and Reuters. The company unveiled Apollo 2, available in both bipedal and wheeled configurations, and said the facility feeds data into Google DeepMind's Gemini Robotics models under an expanded research partnership. Reuters quotes CEO Jeff Cardenas: "We have a factory that produces robots, we also have a factory that produces data." Reuters also reports Apptronik raised $520 million in February at roughly a $5 billion valuation, and that Cardenas expects real production versions of the platform in 2027 and beyond.
For practitioners building embodied AI, the harder problem is rarely model architecture alone, it is generating enough diverse, real-world task data to close the sim-to-real gap. Apptronik's approach treats physical interaction data as a product in its own right, not a byproduct of deploying robots.
What happened
Apptronik announced the opening of Robot Park, a roughly 90,000-square-foot data-collection and training facility in Austin, Texas, in a June 30 press release carried by GlobeNewswire and covered by Reuters. The company also unveiled Apollo 2, its current humanoid platform, available in both bipedal (legged) and wheeled configurations; Reuters reports Apollo 2 has served as a data-collection platform for more than a year. Per the release and Reuters, Robot Park's data feeds Google DeepMind's Gemini Robotics models under an expanded research partnership, and similar data-collection workflows are being deployed at a growing network of Robot Parks at DeepMind and at customers including Mercedes-Benz and GXO. Reuters reports Apptronik raised $520 million in February, valuing the company at about $5 billion, and quotes CEO Jeff Cardenas: "We have a factory that produces robots, we also have a factory that produces data."
Technical context
Physical robot data differs from web-scale text or image corpora because it must capture closed-loop interactions, contact dynamics, varied lighting and object arrangements, and operator interventions. Facilities like Robot Park let teams instrument environments, log high-frequency proprioceptive and vision streams, and run supervised or imitation-learning collection at scale through a mix of teleoperation and autonomous execution. Automate.org quotes Cardenas describing Apollo 2 as a "prototype" and "data collection platform," language meant to distinguish it from a production-deployment unit; Apptronik has said the lessons from Apollo 2 are directly informing its next commercial product, Apollo 3.
For practitioners
Partnerships between robot builders and model labs, illustrated here by Apptronik and Google DeepMind, shorten the feedback loop between hardware deployment and model iteration by giving foundation-model teams task-aligned real-world data rather than relying on simulation transfer alone. Treating data collection as infrastructure changes operational priorities toward tooling, telemetry, and scalable teleoperation workflows, and toward managing deployment heterogeneity across different customer sites.
What to watch
Watch for technical disclosures on the dataset formats, annotation schemas, and training pipelines used with Gemini Robotics, and for independent evaluations of task generalization beyond Apptronik's own facilities. Reuters quotes Cardenas saying Apptronik will "continue to pilot through this year, and then we'll start to see real production versions ... in 2027 and beyond," which frames the near-term horizon for production-scale validation; Reuters also reports Cardenas declined to disclose current deployment numbers, noting only that Apptronik has built "hundreds" of Apollo 2 units.
Key Points
- 1Apptronik opened Robot Park, a 90,000-square-foot Austin facility, to collect real-world task data from its Apollo 2 humanoid robots.
- 2The data feeds Google DeepMind's Gemini Robotics models under an expanded research partnership, tightening the hardware-to-model feedback loop.
- 3Apptronik raised $520 million in February at a roughly $5 billion valuation and expects production-scale deployments in 2027 and beyond.
Scoring Rationale
A material, well-verified investment in physical-AI data infrastructure with a direct, named research link to Google DeepMind's robotics models. Significant for practitioners training embodied agents, but an infrastructure/product milestone rather than a frontier model release, so it stays in the major-not-historic tier.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
