Manan Tomar

Berkeley College, Indian Institute of Technology Madras

Papers

Total Citations

H-Index

About

Manan Tomar is a researcher at the forefront of reinforcement learning (RL) and robotics, with a focus on enabling agents to learn complex behaviors from diverse data sources. His work bridges the gap between large-scale pre-training and practical robotic control, addressing fundamental challenges in skill discovery and curriculum learning. Tomar’s research on "Robotic Offline RL from Internet Videos via Value-Function Learning" (2024, 5 citations) explores how pre-training on internet data can unlock broad generalization in robotic systems, a key step toward scalable RL. He also introduced "Successor Options" (2019, 5 citations), a novel framework for discovering reusable skills by leveraging successor representations, moving beyond traditional bottleneck-based approaches. Additionally, his "MaMiC" framework (2019, 4 citations) proposes a dual curriculum strategy—macro and micro—for solving sparse-reward robotic manipulation tasks, demonstrating how structured guidance can accelerate learning. Though early in his career, Tomar’s contributions are already shaping how researchers think about data-efficient RL and skill transfer, with his work cited in top venues and inspiring new directions in offline RL and hierarchical learning.