Vaidehi Som

Papers

Total Citations

H-Index

About

Vaidehi Som is a researcher at the forefront of embodied AI and robot learning, whose work bridges the critical gap between high-level language understanding and low-level robotic control. Her most impactful contribution, the "LIV" framework (Language-Image Value learning), introduces a unified objective that simultaneously learns vision-language representations and reward functions directly from action-free videos paired with text annotations. By forging a novel theoretical connection between dual reinforcement learning and mutual information contrastive learning, Som enables robots to acquire complex manipulation skills from passive observation, dramatically reducing the need for costly human demonstrations or hand-crafted reward engineering. This pioneering approach, published in 2023 and already garnering 24 citations, has opened new pathways for scalable robot learning from diverse internet data. Som's work is particularly notable for its elegant mathematical formulation, which provides a principled foundation for integrating language, vision, and control. Her research stands as a key reference for students and researchers working at the intersection of computer vision, natural language processing, and reinforcement learning, offering a compelling vision for how robots can learn generalizable skills from the rich, unstructured data available in the modern world.