Rahul Venkatesh

Papers

Total Citations

H-Index

About

Rahul Venkatesh works at the intersection of computer vision, cognitive science, and robotics, with a focus on building machines that perceive the world more like humans do. His research centers on unsupervised segmentation, counterfactual world modeling, and task-general perception—areas where he seeks to move beyond the limitations of task-specific, label-hungry architectures. Venkatesh’s most influential work, "Unsupervised Segmentation in Real-World Images via Spelke Object Inference" (2022, 17 citations), introduces a novel approach that learns static grouping priors from motion self-supervision, drawing on the cognitive science concept of a Spelke object—a unified physical entity that moves coherently. This work tackles the challenging problem of category-agnostic segmentation without labeled data. In a follow-up paper (2023, 9 citations), Venkatesh proposes unifying machine vision through counterfactual world modeling, arguing that robust, task-general perception—critical for fields like robotics—has been held back by fragmented, task-specific architectures. His contributions offer a path toward more flexible, human-like visual systems, with potential to transform how machines interact with dynamic, real-world environments.