Shagun Sodhani

Papers

Total Citations

H-Index

About

Shagun Sodhani is a leading researcher at the intersection of reinforcement learning (RL), robotics, and representation learning, whose work is shaping how machines acquire generalizable skills. His most impactful contributions tackle two foundational challenges: learning universal reward functions and mastering multiple tasks simultaneously. In his highly cited work "VIP: Towards Universal Visual Reward and Representation via Value-Implicit Pre-Training" (2022, 35+ citations), Sodhani pioneered a method to learn a single, scalable reward and visual representation from diverse, offline human videos—dramatically reducing the need for costly, in-domain robot data. This approach enables robots to understand and replicate complex manipulation skills by leveraging abundant human demonstrations. Complementing this, his paper "Multi-Task Reinforcement Learning with Context-based Representations" (2021, 20+ citations) introduced a framework that improves multi-task RL by learning context-aware representations, allowing agents to share knowledge across tasks more effectively than traditional shared-parameter methods. Together, these works demonstrate Sodhani’s ability to bridge theory and practice, offering scalable solutions for real-world robotics. His research is vital for students and engineers aiming to build agents that learn efficiently from limited data and adapt across diverse environments.