Deepanshi Srivastava

Master's

Computer Science

Cornell University

Research Assistant

- Designed and implemented innovative computer vision modules for state identification and manipulation of deformable - Engineered deep learning pipelines (UNets, CNNs with custom loss functions, YOLO) to enhance localization and feature-specific parameter regression, achieving 96.75% accuracy, and contributing to the development of scalable code - Developed a computer vision-based perception module for robot-assisted dressing of quadriplegic patients, focusing on state identification and manipulation of deformable fabric, and grasp acquisition, deployed on a 7 DoF Franka Robot - Approach involved generating several deep learning pipelines using UNets forregion segmentation, CNN with custom loss forfeature-specific parameterregression, custom image augmentations, and YOLO for robust localization and classification of state-specific high-level features along with robust mathematical planning and depth estimation. Accuracy: 96.75%.