Research

My research focuses on machine learning methods for embodied intelligence, with an emphasis on generative models, 3D scene understanding, and multimodal foundation models. My work enables robots and agents to perceive and act in partially observed environments. My recent work introduces a framework for real-time 3D occupancy prediction in mobile robots, bridging generative AI with robotic perception and map-building.

I am broadly interested in foundation models for vision, language, and action; training and deployment of large models in embodied settings; and developing algorithms which improve efficiency and reliability of autonomous systems.

Robust Robotic Exploration and Mapping Using Generative Occupancy Map Synthesis

In Review at Autonomous Robots Check out the pre-print manuscript here.

Online Diffusion-Based 3D Occupancy Prediction at the Frontier with Probabilistic Map Reconciliation

Paper Webpage

Accepted to ICRA 2025: Check out the pre-print manuscript here.

SceneSense: Diffusion Models for 3D Occupancy Synthesis from Partial Observation

Paper Webpage

Accepted to IROS 2024: Read our pre-print manuscript here. Read the published version on IEEE Xplore