Kasra Darvish
Papers
5
Total Citations
24
H-Index
4
About
Kasra Darvish is a researcher advancing the frontier of grounded language learning for human-robot interaction. His work sits at the intersection of natural language processing, computer vision, and robotics, tackling the fundamental challenge of teaching machines to understand language in the context of the physical world. Darvish’s key contributions include developing practical cross-modal manifold alignment techniques that use triplet loss to create consistent, multi-modal embeddings of real-world objects from RGB-depth data. He has also pioneered the use of virtual reality environments to generate rich, scalable training data for grounded language acquisition, addressing the critical barrier of data scarcity in real-world deployments. His work on category-free learning of object attributes demonstrates a path toward more flexible, open-world robotic understanding. With multiple papers accumulating citations in the single digits—a strong start for early-career work—Darvish’s research is gaining traction. Notably, his creation of a spoken language dataset for speech-based grounded learning provides a vital resource for the community, pushing toward more natural, intuitive interfaces between humans and robots.
Research Focus
Key Achievements
Top Papers
- 1
- 2Towards Making Virtual Human-Robot Interaction a Reality6 citations · 2021
- 3
- 4
- 5Practical Cross-modal Manifold Alignment for Grounded Language2 citations · 2020