Can a humanoid robot improve simply by working every day? A modular platform is collecting real-world data to train future AI systems.

Apptronik has launched Apollo 2, the latest version of its AI-powered humanoid robot platform, designed to accelerate the development of embodied artificial intelligence through continuous real-world learning. Available in both bipedal and wheeled configurations, the platform is intended to gather operational data across industrial environments while supporting the company’s research collaboration with Google DeepMind.
Unlike systems trained primarily in simulation, Apollo 2 is built to learn from practical deployments. The robots collect data while performing tasks in logistics, manufacturing, retail and other customer environments, creating datasets that are used to refine Google DeepMind’s Gemini Robotics foundation models. According to the company, this continuous learning approach is aimed at improving the reliability and adaptability of future humanoid robots.
To support large-scale data collection, Apollo 2 operates across Apptronik’s Robot Park network, including its newly expanded Austin facility and customer locations such as Mercedes-Benz, GXO and Google DeepMind. The platform combines teleoperation, autonomous execution and high-fidelity physics simulations to generate training data while enabling engineers to evaluate robot behaviour in real operating conditions. Its modular design allows the same humanoid platform to be deployed either as a wheeled robot compatible with existing industrial safety standards or as a bipedal system for navigating more complex environments.
The company says experience gained through Apollo 2 will directly contribute to the development of its next commercial humanoid platform, Apollo 3, with the objective of improving out-of-the-box embodied intelligence and supporting broader industrial deployment.
“The industry has spent years showing what robots can do in demos. We’re focused on what they can do every day on the job. What we’re building is a continuous learning loop with the Google DeepMind Robotics team: robots working, collecting data and improving with every cycle, in real environments, on real tasks. Robot Park enables the data collection that is fuel for that, and Apollo 2 is the machine that makes it possible. That’s how you move from early prototypes to real, deployable humanoid robots,” says Jeff Cardenas, Chief Executive Officer and Co-founder of Apptronik.
Click here for the official announcement.



