Shamit Lal

Papers

Total Citations

H-Index

About

Shamit Lal is a researcher advancing the frontier of 3D scene understanding and interactive simulation. His work centers on developing neural representations that enable machines to perceive, predict, and interact with dynamic environments from visual data. In his most-cited paper, "3D-OES: Viewpoint-Invariant Object-Factorized Environment Simulators" (2020, 13 citations), Lal proposed a groundbreaking action-conditioned dynamics model that predicts scene changes caused by object and agent interactions. The key innovation lies in operating within a viewpoint-invariant 3D neural scene representation space, inferred directly from RGB-D videos. By factorizing objects so they do not interfere with one another, the model achieves robust appearance persistence and disentangled reasoning about individual scene elements. This work has significant implications for robotics, embodied AI, and simulation—enabling agents to learn from visual observations and anticipate the consequences of their actions across different viewpoints. Lal’s contributions help bridge the gap between raw perception and structured world models, offering a path toward more generalizable and physically grounded AI systems.