Papers
19
Total Citations
720
H-Index
9
About
Sudeep Dasari is a robotics researcher whose work sits at the intersection of deep learning, reinforcement learning, and generalizable robotic manipulation. He has made significant contributions to model-based deep reinforcement learning for robotic control, most notably through "Visual Foresight" (2018, 264 citations), which demonstrated that robots could learn complex visuomotor skills from raw sensory inputs in real-world settings. His early work also pioneered one-shot imitation learning from human video demonstrations using domain-adaptive meta-learning (113 citations), a remarkably ambitious step toward human-like skill acquisition in robots. Dasari has consistently championed large-scale, data-driven approaches to robotic learning. His contributions to RoboNet and the landmark DROID dataset (108 citations) reflect a sustained effort to build the diverse, large-scale training corpora that generalizable robot policies require. His co-authorship on Octo (2024, 66 citations), an open-source generalist robot policy, underscores his commitment to accessible, community-driven robotics research. More recent work on dexterous manipulation and human-affordance-based pre-training further highlights his evolving focus on building robots that can learn efficiently and transfer broadly—an increasingly vital frontier as robotics moves toward real-world deployment.
Research Focus
Key Achievements
Top Papers
- 1
- 2One-Shot Imitation from Observing Humans via Domain-Adaptive Meta-Learning113 citations · 2018
- 3DROID: A Large-Scale In-The-Wild Robot Manipulation Dataset108 citations · 2024
- 4Octo: An Open-Source Generalist Robot Policy66 citations · 2024
- 5
- 6
- 7Model-Based Visual Planning with Self-Supervised Functional Distances17 citations · 2020
- 8RoboNet: Large-Scale Multi-Robot Learning16 citations · 2019
- 9
- 10HRP: Human affordances for Robotic Pre-training8 citations · 2024