Image of Ayush Agrawal

Budding Researcher
Cognitive AI

CV

Hello I'm Ayush! A researcher passionate to equip embodied agents with physical commonsense reasoning capabilities. Currently, I am working on developing useful language based abstractions for open world navigation with Dr. David Hsu. Through my previous works I've worked on commonsense reasoning and scene understanding in embodied agents.

Before this, I worked on commonsense based object selection while being advised by Dr. Dianbo Liu and Dr. Anirudh Goyal. Even before, I worked on developing commonsense based object navigation techniques during my time as a research assistant at Robotics Research Center, IIIT Hyderabad. I've obtained my bachelors in B.E Electronics & Instrumentation Engineering from BITS Pilani, Pilani in 2022.

During my undergraduate days, I was extremely passionate about insect behavior and collective intelligence. My bachelors thesis was based on developing Honey Bee Vision inspired Obstacle Avoidance Algorithms under the guidance of Dr. Sridhar Ravi

more

Education

  • B.E. in Electronics & Instrumentation, 2022
    BITS Pilani, Pilani

News

Publications

image of submitted paper
Physical Reasoning and Object Planning for Household Embodied Agents
Ayush Agrawal, Raghav Prabhakar, Anirudh Goyal, Dianbo Liu
TMLR 2024|OpenReview|code|dataset

Demystifying the decision making process behind choosing an object for task completion, we develop a 3 step architecture and curate datasets to power future research in this domain. Further, we evaluate various LLM baselines and report the findings.


Image of Ayush Agrawal
Sequence Agnostic MultiON
Nandiraju Gireesh*, Ayush A*, Ahana Datta* ... Madhava Krishna
ICRA 2023 |arxiv|video|blog

You are already in a kitchen, and tasked to find a fridge. Would you search for it in the current area or other places in house? We train a RL policy based off semantic relationship between static objects to generate efficient long term goals to enable quick retreival of a list of objects.


Image of Ayush Agrawal
CLIPGraphs: Multimodal Graph Networks to Infer Object-Room Affinities
Ayush A*, Raghav Arora*, ... Madhava Krishna
RO-MAN 2023 |arxiv|code|page

Leveraging upon the knowledge that we humans have highly developed Object-Utility and Room-Utility relationships; we generate human commonsense aligned latent embeddings useful for varius Embodied AI tasks. We do this by developing a Graph Neural Network by processing Human Preference datasets and Foundation Model Features.