Embodied AI development focuses on industrial use and simulation testing

The development of embodied AI, also known as physical AI, is expected to see faster implementation in factories and logistics centers than in domestic environments. Experts note that while industrial robots have performed tasks like welding for decades, the current challenge lies in enabling AI to interpret surroundings and respond to unprogrammed situations without constant human supervision.

To be effective, these robots must reliably navigate the physical world, anticipating consequences such as the fragility of objects or the impact of their movements. Research into ‘world models’ aims to help these machines learn how the physical environment behaves.

In a practical application of these concepts, Nvidia has introduced Microduck, a small, $399 robot designed for experimentation and robotics development. The device can perform movements such as sitting, sleeping, and flipping. Microduck serves as a tool for developers to test how AI can transfer learned behaviors from virtual simulations to the physical world.