Papers
18
Total Citations
262
H-Index
8
About
Homanga Bharadhwaj is a robotics and machine learning researcher whose work sits at the intersection of robot learning, imitation learning, and reinforcement learning, with a particular focus on enabling robots to generalize across diverse real-world manipulation tasks. His most cited work, "RoboAgent" (2024, 70 citations), addresses one of the field's central challenges: building a single, versatile robot agent capable of manipulating arbitrary objects despite the scarcity of large-scale robotics datasets, leveraging semantic augmentations and action chunking to dramatically improve generalization and efficiency. Complementing this, his "Learning by Watching" framework (2021, 43 citations) pioneered approaches for extracting manipulation skills directly from human videos, reducing the need for costly hand-engineered demonstrations. His "Track2Act" work (2024, 34 citations) further advances this vision by predicting point tracks from internet videos to enable generalizable robot control. Beyond manipulation, Bharadhwaj has contributed to continual model-based reinforcement learning using hypernetworks and explored scalable multi-task imitation learning through the CACTI framework. With over 200 cumulative citations spanning navigation, affordance prediction, and deep architecture design, his research consistently pushes toward more scalable, data-efficient, and generalizable robot learning systems.
Research Focus
Key Achievements
Top Papers
- 1
- 2
- 3
- 4Continual Model-Based Reinforcement Learning with Hypernetworks20 citations · 2021
- 5
- 6
- 7D2RL: Deep Dense Architectures in Reinforcement Learning13 citations · 2020
- 8Visual Affordance Prediction for Guiding Robot Exploration11 citations · 2023
- 9
- 10Continual Model-Based Reinforcement Learning with Hypernetworks6 citations · 2020