Shashank Hegde

A single JEPA latent space unifies state representation, inverse dynamics, and policy prediction, outperforming prior reconstruction-based world action models in state alignment and achieving competitive closed-loop control.

Training paradigms dictate orientation selectivity in vision transformers, revealing that early layers excel at encoding low-level features while deeper layers transition to semantic representations.

CLAE enables real-time behavior adaptation in robots without retraining, allowing for unprecedented flexibility in dynamic environments.