Shanghai based Fysics AI has launched Fysiverse, a new generation physics based world model that integrates real world physical laws directly into its code. This approach contrasts with the data driven methods favored by American tech giants like OpenAI and Meta Platforms.
In an announcement on its WeChat social media account, Fysics AI described Fysiverse as “a new generation physics-based world model that adheres to real world physical laws.” The company, founded by former Nvidia senior manager Zhang Lihua, claims the new paradigm could address common issues in current world models such as “physical illusions, reasoning failures, and breakdowns in non standard scenarios.”
World models are crucial for creating content and training robots or self driving technology. Currently, there are three major paradigms dominating this field:
1. Black Box Approach: This method allows the model to construct its own rules of the world without knowledge of physics. Meta’s V-JEPA series, which stands for video joint embedding predictive architecture, exemplifies this approach through a self supervised learning process.
2. Three Dimensional Modeling: In this paradigm, the model learns, predicts, and simulates the geometry and dynamics of real environments from text, images, videos, or coarse 3D elements.
Fysics AI’s Fysiverse represents a departure from these existing approaches by embedding explicit physical laws directly into its code. This unique feature could provide more accurate and reliable simulations, potentially offering significant advantages in applications such as robotics and autonomous vehicles where precise physical interactions are critical.
The launch of Fysiverse marks a shift in the physics based world modeling landscape, presenting a new solution to longstanding challenges faced by current models.
Thinking about building an AI product?
Get in Touch