Artificial Brains

We build embodied models based on Spiking Neural Networks for general robotic intelligence.

Instead of mapping the world through trillions of data points (e.g., World Models or VLAs), our algorithm allows the robot to develop an understanding of its own embodiment: where its body ends and the world begins, and how its actions impact the world through continuous cycles of action and consequence.

Our approach bridges the gap between training and inference in physical intelligence, providing early evidence of real-time learning and adaptation, with advantages in energy consumption, computing and data requirements.