Thanuja Dharmasiri

Papers

Total Citations

H-Index

About

Thanuja Dharmasiri is a researcher at the forefront of 3D scene understanding for autonomous robotics, specializing in deep learning architectures that extract rich geometric information from single RGB images. Her most impactful work introduces a novel convolutional neural network framework capable of jointly predicting depth, surface normals, and surface curvature from a single RGB image—a breakthrough that enables robots to perceive complex 3D structures without relying on expensive sensors like LiDAR. This pioneering approach, detailed in her highly cited 2017 paper (31 citations), addresses a critical challenge in robotics: understanding scene geometry from minimal input. Building on this, Dharmasiri advanced the field by developing a real-time system that simultaneously performs semantic segmentation and depth estimation using asymmetric annotations (2019, 16 citations), tackling the practical hurdles of deploying multi-task deep learning models on resource-constrained robotic hardware. Her work demonstrates that a single, efficient model can replace multiple specialized networks, significantly reducing computational overhead while maintaining accuracy. With over 60 combined citations across her key publications, Dharmasiri’s contributions are shaping the next generation of autonomous systems that can navigate and interact with their environments more intelligently and efficiently.