Insider Brief
- Joyx and Hitch Interactive formed a strategic partnership to develop and deploy embodied AI systems for commercial use by combining robot-training data with standardized robotics hardware and software.
- Joyx will provide capture, spatial reconstruction and verification of human and robot demonstration data through its real-to-real pipeline, while Hitch Interactive will contribute its Hitch Open robotics platform, control interfaces, open-source software and operating environments.
- The companies aim to build complete systems for specific commercial scenarios, using real-world video and sensor data that is reconstructed into metric 3D, adapted to a robot’s physical constraints and returned to the robot as training data.
Joyx and Hitch Interactive have formed a strategic partnership to develop and deploy embodied AI systems for commercial use.
Joyx said it will provide the data side of the partnership, including capture, spatial reconstruction and verification of human and robot demonstration data. Hitch Interactive will provide access to its Hitch Open robotics platform, along with standardized hardware and control interfaces, an open-source software stack and the environments where the robots operate.
The U.S. companies plan to use those pieces to build complete systems for specific commercial scenarios rather than offering separate data, software or hardware components.
“Physical AI is gated by one thing: knowledge of the physical world that a machine can actually act on,” Joyx founder and CEO Tong Zhang said in the announcement. “Raw video of the real world is abundant and nearly free; verified, metric-accurate understanding built from it is scarce and valuable. That is what Joyx builds — spatial intelligence from ordinary video, carried through our real-to-real pipeline all the way onto real robots. Hitch Interactive gives us the interfaces and the fleet to prove it where it matters: working systems in real vertical scenarios.”
The process begins with real-world video and sensor data, which Joyx reconstructs into a metric 3D representation. That data is then adapted to the physical structure and movement limits of a specific robot, checked against real-world constraints and returned to the robot as training data.
Joyx noted the approach is intended to address limitations in two common methods for training robots. Teleoperation, in which a human directly controls a robot, can generate accurate demonstrations but is difficult to scale across large numbers of tasks and environments. Simulation can generate larger amounts of training data but can introduce differences between virtual environments and how robots behave in the physical world.
Joyx’s approach instead begins and ends with real-world data, while Hitch Interactive supplies the robot platforms and operating environments needed to put that data into use.
“The gap in physical AI is increasingly not about hardware — it is data that robots can actually learn from,” added Hitch Interactive co-founder Allen Yang. “Joyx holds a genuinely unusual position here, with a solution set built to close exactly that gap. Pairing it with open, standardized interfaces means a capability proven on one platform can move quickly to many, which is what the field needs.”
