LaGSplat framework infers physics-governed simulations from video

Researchers have developed LaGSplat, a novel framework that infers physics-governed interactive simulations from monocular video. This system allows users to apply external forces to filmed objects, enabling real-time rendering of their responses in 2D or 3D. LaGSplat utilizes a latent state that acts as both the generalized coordinate for a learned dissipative Lagrangian and a conditioning variable for a Gaussian Splatting decoder, enabling it to translate image-space forces into generalized forces for the equations of motion. AI

IMPACT Enables more realistic and interactive simulations from video data, potentially impacting fields like robotics and augmented reality.

RANK_REASON The cluster contains a research paper detailing a new framework for inferring physics-governed simulations from video. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• CNN
• Euler–Lagrange equation
• Gaussian splatting
• Lagrange function
• LaGSplat
• Latent Lagrangian Gaussian Splatting
• Nerf

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →