Nvidia Brute-Force Bubble: Why 90% of Physics AI Compute Is a Mathematical Waste (github.com)

A new project has emerged aiming to address the significant challenge known as the "5000W Compute Wall" in physics-based AI simulations. Traditional physics engines, like PhysX and MuJoCo, struggle with efficiency due to their reliance on discrete time-stepping methods. This development introduces a novel approach using octonions, a largely unexplored mathematical construct in robotics, as a 256-bit algebraic container that internalizes time, potentially streamlining simulations and enhancing computational efficiency. The implications of this work are profound for the AI/ML community, as it shifts from conventional brute-force methods to a more algebraically efficient model. By utilizing the non-associative properties of octonions, the new method can accurately reflect the dynamics of multiple collision events without reverting to traditional constraint solvers. This advancement promises to lock collision sequences algebraically, ensuring physical realism by preventing causality violations during simulations. The researchers are seeking collaboration with NVIDIA to validate their findings and explore dedicated hardware solutions, aiming to redefine the physics engine framework essential for robotics applications.