Soubarna Banik

About

Soubarna Banik’s research sits at the intersection of computer vision, deep reinforcement learning, and robotics, with a particular focus on integrating structural knowledge into learning systems. Her most cited work, “Graph Neural Networks for Relational Inductive Bias in Vision-based Deep Reinforcement Learning of Robot Control” (2022), addresses a critical limitation in standard reinforcement learning algorithms: their inability to exploit the inherent relational structure of robotic tasks. By introducing graph neural networks to encode relational inductive biases directly into vision-based policies, Banik demonstrates how prior knowledge of object interactions and spatial relationships can dramatically improve learning efficiency and policy generalization. This contribution is especially significant for real-world robotics, where tasks often involve manipulating multiple objects with clear relational constraints. Though early in her career, with 2 citations on this flagship paper, Banik’s work has already been recognized for its innovative bridging of graph representation learning and robot control. Her research offers a promising path toward more sample-efficient, structurally aware AI systems capable of reasoning about the physical world in ways that mirror human intuition.

Research Focus

Key Achievements

Top Papers

1. 1
   Graph Neural Networks for Relational Inductive Bias in Vision-based Deep Reinforcement Learning of Robot Control

   2 citations · 2022

Key Collaborators

Contact & Links