Meta Unveils New AI Model for Understanding Physical Laws

• Meta has developed the V-JEPA 2 model for predicting physical changes in the environment.
• The model allows robots to adapt to new situations without the need for large amounts of training data.
• The technology has the potential to improve efficiency in logistics and manufacturing by understanding causal relationships.

This was reported by CNET, VentureBeat.

The company Meta has introduced a new artificial intelligence model that could change the way robots interact with the physical world. The new model, known as V-JEPA 2, enables robots to predict changes in their environment and make more informed decisions.

This open model, designed to train AI to understand concepts such as gravity and object permanence, could be beneficial for autonomous vehicles and robots, as it does not require vast amounts of training data to adapt to new situations. Meta hopes that the development of such systems will contribute to improving people's quality of life.

Key Features of the V-JEPA 2 Model

The V-JEPA 2 model is trained on video and physical interactions, allowing it to understand what is happening in a scene, predict changes in the scene based on actions, and plan sequences of actions to achieve goals. This architecture reduces data processing costs by focusing on high-level features of the scene, such as object positions, rather than detailed texture representations.

V-JEPA 2 undergoes training in two stages. First, it learns physics by observing millions of hours of unlabeled video, and then it refines its knowledge based on a specialized dataset to link actions with physical outcomes.

Impact on Business and Industry

This model provides the opportunity to develop adaptive robots that can operate in environments with changing products and warehouse layouts without the need for significant reprogramming. This could be beneficial in fields such as logistics and manufacturing, where understanding causal relationships is crucial.

Meta plans to create a broad community around this technology by providing training code and hoping for progress in developing models that can fundamentally change the way AI interacts with the physical world.

Теги: Технології/ШІ