Jyothish Pari

Papers

Total Citations

128

H-Index

About

Jyothish Pari is a rising star in robot learning, whose work tackles the core challenges of data efficiency and generalization in imitation learning. His research focuses on enabling robots to learn complex manipulation skills from minimal human demonstrations, bridging the gap between offline training and real-world deployment. Pari’s most influential paper, “The Surprising Effectiveness of Representation Learning for Visual Imitation” (2022, 82 citations), demonstrated how pre-trained visual representations can dramatically reduce the need for hundreds of diverse demonstrations, making imitation learning far more practical. He extended this line of work with “Teach a Robot to FISH: Versatile Imitation from One Minute of Demonstrations” (2023, 30 citations), showing that robots can acquire robust, generalizable skills from just a single minute of human video. Pari also co-led the “Train Offline, Test Online” benchmark (2023, 16 citations), a landmark effort to standardize robot learning evaluation across labs by using a shared, low-cost robot platform. This benchmark directly addresses the reproducibility crisis in robotics, enabling more labs to participate in meaningful research. Through his focus on data-efficient, representation-driven methods, Pari is helping to democratize robot learning and accelerate the path toward general-purpose robots.