Nihal Soans

Papers

Total Citations

H-Index

About

Nihal Soans is a robotics researcher whose work sits at the intersection of computer vision and imitation learning, with a primary focus on enabling robots to learn complex tasks simply by watching humans. His key research area is Learning from Observation (LfO) and Learning from Demonstration (LfD), where he addresses the fundamental challenge of how a robot can automatically decompose raw sensory data into meaningful sequences of state-action pairs. Soans’s major contribution is the development of SA-Net, a deep neural network architecture designed for robust state-action recognition from RGB-D video streams. This work, first published in 2020, has garnered 24 citations and represents a significant step toward making robot learning more autonomous and less reliant on manual programming or teleoperation. By tackling the "correspondence problem" in imitation learning—how a robot maps observed human actions to its own embodiment—Soans’s research helps bridge the gap between raw perception and actionable robotic knowledge. His earlier work on trajectory recognition (2019) laid the groundwork for this approach, and together, these contributions advance the vision of robots that can learn new skills through passive observation, a paradigm with profound implications for accessible, intuitive human-robot interaction.