1X, the robotics firm responsible for the Neo humanoid robot, has launched an AI model designed to comprehend real-world physics, enabling its robots to acquire new skills autonomously.
Dubbed the 1X World Model, this physics-driven system empowers Neo robots with fresh capabilities by integrating video inputs and prompts. According to 1X, the video component specifically allows Neo to grasp tasks it hasn’t been explicitly trained for.
This introduction coincides with 1X’s preparations to deploy its Neo humanoids into consumer homes. Preorders for these robots commenced in October, with deliveries projected for this year. While a 1X spokesperson confirmed that preorders surpassed expectations, they did not provide specific shipping dates or order figures.
In a statement, Bernt Børnich, founder and CEO of 1X, stated, “Following years dedicated to refining our world model and designing Neo to be as human-like as possible, Neo is now capable of learning from vast internet video data and applying that understanding to physical environments. Its capacity to convert any prompt into novel actions, even without previous instances, signifies the beginning of Neo’s potential to self-learn and excel at virtually any requested task.”
Claiming the robot can convert any prompt into a new action is a bold assertion and not entirely precise; for example, instructing Neo to drive a car won’t instantly enable it to parallel park. Nevertheless, a degree of learning is indeed occurring.
A company spokesperson clarified that 1X does not suggest the world model immediately enables current Neo bots to perform new tasks simply by capturing video and receiving prompts. Rather, the robot collects video data associated with particular prompts and transmits it back to the world model. This model then informs the robot network, enhancing their comprehension of the physical world and their practical expertise.
Furthermore, this provides users with insight into Neo’s intended behavior or reaction to a given prompt. Such behavioral data could be instrumental for 1X in training these models, ultimately allowing robots to respond to prompts for tasks they have never encountered previously.