Bridging the paradigm gap between Vision-Language Models (VLMs) and real-world physical interactions. We focus on multi-agent multimodal robotics, adaptive hardware design, and resilient embodied AI systems.
The Physical Intelligence Lab (PILab) at IIT Kharagpur approaches robotics from an applied, experimental "builder" perspective. We construct tight hardware-software integrations capable of complex open-ended multi-agent tasks across terrestrial, aerial, and aquatic configurations.
Fusing deep semantic vision-language representation spaces with downstream robotic feature spaces for adaptive control.
Developing scalable, decentralized communication and optimization frameworks for cooperative autonomous task executions.
Constructing high-fidelity campus-scale point clouds and submap-fused 3D Gaussian Splatting configurations for intelligent navigation.
Deploying tiered models connecting deep high-level reasoning to agile low-level micro-controller frequency actuation loop execution.
Pushing the frontier of Physical AI through coordinated hardware-software co-design initiatives.
Developing fine-tuned Vision-Language models by optimizing specialized custom loss function configurations. Our focus includes decreasing text/caption dimensionality down to functional atomic primitives to accelerate continuous video-sequence sequence reasoning.
Generating dense, scalable digital twins using continuous 3D Gaussian Splatting and semantic SLAM pipelines. We implement specialized submap fusion routines to deploy real-time context-aware spatial navigation directly inside high-density point clouds.
Constructing multi-tier embodied systems where natural language instructions accurately map to unified multi-stage spatial trajectories and robust precision micro-actuation manipulation hooks.
Our physical testbeds are designed, modeled, and manufactured in-house. We heavily focus on creating customized, efficient kinetic configurations over rigid commercial out-of-the-box hardware templates.
Core scientific and mechanical research output from our team.
H. Saha, F. Fotouhi, Q. Liu, S. Sarkar
H. Saha, A. R. Florita, G. P. Henze, S. Sarkar
Dr. Dayal Parhi, Dr. Chinmaya Sahu, Dr. Homagni Saha (Patent No. 529812)
Leading interdisciplinary initiatives at the nexus of mechanics and computer vision science.
Dr. Saha heads the Physical Intelligence Lab at the Centre for Interdisciplinary and Convergent Technologies, IIT Kharagpur. He completed his Ph.D. in Mechanical Engineering and Computer Science at Iowa State University, followed by industry positions as a Lead Machine Learning Scientist and Computer Vision Engineer.
We are actively seeking motivated M.Tech, Ph.D., and undergraduate research interns passionate about Physical AI, 3D printing custom robotics, and Vision-Language models.