UbiNet: Ubiquitous Networked Systems Lab

19/08/2026

𝐈𝐭 𝐰𝐚𝐬 𝐚 𝐩𝐫𝐢𝐯𝐢𝐥𝐞𝐠𝐞 𝐭𝐨 𝐰𝐞𝐥𝐜𝐨𝐦𝐞 𝐬𝐨𝐦𝐞 𝐨𝐟 𝐨𝐮𝐫 𝐝𝐢𝐬𝐭𝐢𝐧𝐠𝐮𝐢𝐬𝐡𝐞𝐝 𝐚𝐥𝐮𝐦𝐧𝐢 𝐭𝐨 𝐨𝐮𝐫 𝐥𝐚𝐛𝐨𝐫𝐚𝐭𝐨𝐫𝐲 𝐝𝐮𝐫𝐢𝐧𝐠 𝐭𝐡𝐞 𝟕𝟓𝐭𝐡 𝐅𝐨𝐮𝐧𝐝𝐚𝐭𝐢𝐨𝐧 𝐃𝐚𝐲 𝐜𝐞𝐥𝐞𝐛𝐫𝐚𝐭𝐢𝐨𝐧𝐬 𝐨𝐟 𝐈𝐧𝐝𝐢𝐚𝐧 𝐈𝐧𝐬𝐭𝐢𝐭𝐮𝐭𝐞 𝐨𝐟 𝐓𝐞𝐜𝐡𝐧𝐨𝐥𝐨𝐠𝐲, 𝐊𝐡𝐚𝐫𝐚𝐠𝐩𝐮𝐫.

We were delighted to host several of our distinguished alumni including Dr. Anand Deshpande, Founder, Chairman & Managing Director of Persistent Systems; Dr. Indradeep Ghosh, CEO of Fujitsu Research of America; Mr. Daibashish Gangopadhyay, Principal Silicon Architect, Display Technologies at Apple Inc., and others at our lab. During their visit, our students and researchers demonstrated several AI-driven smart sensing solutions being developed in the lab. The demonstrations showcased how advances in artificial intelligence, sensing, and ubiquitous computing can be brought together to address challenging real-world problems.

It was particularly rewarding to interact with our distinguished alumni, discuss our ongoing research directions, and receive their valuable perspectives and feedback. Such interactions are immensely encouraging for our students and researchers and help us think more ambitiously about translating research innovations into meaningful real-world impact.

The visit made the Institute’s 75th Foundation Day even more special for us -- a wonderful occasion to celebrate IIT Kharagpur’s legacy while sharing a glimpse of the research and innovation shaping its future. We sincerely thank the visitors for taking the time to visit our lab and interact with our team.

16/08/2026

UbiNet Lab (CSE, IIT Kharagpur) showcased three translational research systems at the IIT Kharagpur Platinum Jubilee Technology Demonstration, Kharagpur Research Park, New Town Kolkata -- an industry-connect event aimed at strengthening the lab-to-market pipeline for academic research.

Prasenjit Karmakar and Arup Sau presented the demos and engaged with attendees throughout the session.

🔹 𝐃𝐀𝐋𝐓𝐎𝐍 -- 𝐈𝐨𝐓 𝐈𝐧𝐝𝐨𝐨𝐫 𝐀𝐢𝐫-𝐐𝐮𝐚𝐥𝐢𝐭𝐲 𝐏𝐥𝐚𝐭𝐟𝐨𝐫𝐦

Captures temperature, humidity, and eight pollutants (CO2, VOC, PM1, PM2.5, PM10, NO2, C2H5OH, CO), paired with real-time activity labels from a speech-to-text app to contextualize readings. This links pollution patterns to specific activities, enabling source/activity classification and informing control policies for ACs, exhausts, and purifiers. Built on the largest indoor pollution dataset to date from India (30+ sites, 6 months, 89.1M samples) and introduces the Healthy Home Index. Published at ACM JCSS 2024, NeurIPS 2024.

🔹 𝐂𝐨𝐖𝐞𝐚𝐫 -- 𝐌𝐚𝐤𝐢𝐧𝐠 𝐈𝐧𝐯𝐢𝐬𝐢𝐛𝐥𝐞 𝐂𝐎𝟐 𝐈𝐧𝐭𝐞𝐫𝐚𝐜𝐭𝐚𝐛𝐥𝐞 𝐰𝐢𝐭𝐡 𝐀𝐑

A wrist-worn CO2 sensor paired with a smartphone AR app that renders pollution as color-coded, interactive bubbles, turning abstract air-quality data into something people can see and act on. A gamified AR mechanic nudges users toward ventilation actions in real time. The system was live demonstrated with AR/VR headsets. Published at ACM CHI 2026.

🔹 𝐑𝐚𝐝𝐚𝐫𝐓𝐫𝐚𝐜𝐤 -- 𝐀𝐮𝐭𝐨 𝐍𝐚𝐯𝐢𝐠𝐚𝐭𝐢𝐨𝐧 𝐮𝐧𝐝𝐞𝐫 𝐙𝐞𝐫𝐨 𝐕𝐢𝐬𝐢𝐨𝐧

A cross-platform ego-motion estimation solution using a single COTS mmWave radar. A fourth-order kinematics-based analytic phase equation drives accurate ego-speed estimation, while Doppler-based speed profiling separates moving from stationary objects in cluttered scenes. Runs at ~0.29s latency on a Jetson Nano, 85% faster than SOTA (Radarize). Built for firefighters, mine workers, and other GNSS/vision-denied environments. Published at SMARTCOMP 2025.

Across these systems, UbiNet's work spans four core themes: Sensing, Computer-Human Interaction, Distributed Systems, and Networking, perceiving the physical world through multi-modal sensors, designing intuitive and assistive interfaces, building resilient distributed systems, and connecting it all through high-performance networks.

Great to connect with industry partners exploring real-world deployment of ubiquitous sensing research.

Thanks to Google Research, Anusandhan National Research Foundation (ANRF), IIT Kharagpur AI4ICPS I Hub Foundation to support our research.

05/07/2026

🎓 Proud moment for at IIT Kharagpur!

Five of our brilliant researchers convocated at the 2026 Convocation of IIT Kharagpur. Huge congratulations to each of them for their outstanding contributions to ubiquitous computing, networked systems, and security research:

🔹 Dr. Sugandh Pargal (PhD) — "Driving Behavior Under Diverse Micro-cultures: Developing Ubiquitous Systems for Safe Driving"

🔹 Dr. Utkalika Satapathy (PhD) — "Provenance and Observability Frameworks for Secure Distributed Microservices"

🔹 Dr. Tirthankar Sengupta (PhD) — "Auditability, Access Control and Redactability in Interoperable Blockchains"

🔹 Dr. Argha Sen (PhD) — "Advancing Next-Generation mmWave Sensing Across Automotive and Human-Centric Environments"

🔹 Ms. Neha Chowdhary (MS) — "Automated Policy Translation and Enforcement for Cloud-Native Deployments"

Each of these theses reflects years of rigorous research, late nights, and a genuine passion for solving hard, real-world problems, from road safety and secure microservices to blockchain auditability, mmWave sensing, and cloud policy enforcement.

We are incredibly proud of what you've achieved and excited to see where your journeys take you next. Congratulations, Doctors and graduates! 🎉

27/06/2026

🎓 Thank you to our 2026 summer interns at UbiNet!

This summer, the UbiNet: Ubiquitous Networked Systems Lab at IIT Kharagpur had the pleasure of hosting a talented cohort of interns, who have now wrapped up their projects with final presentations.

Over the past few months, they explored a wide range of problems across our research areas, from 5G network performance monitoring and mmWave radar sensing to WiFi-based human activity recognition and distributed observability in edge computing. Their curiosity, hard work, and fresh ideas brought real energy to the lab.

A big thank you to each of them for being part of the UbiNet team:

Satabarto Sarkar, Sagar Patra, Kanishak Mittal, Sneha Guha, Souherdya Sarkar, Puneeth Pemmasani, Tamaghna Kundu, Saptarshi Sarkar, Sayantan Das

We are proud of everything you accomplished and wish you the very best for the road ahead. Stay in touch, and keep building!

26/06/2026

🎓 Congratulations, 𝐃𝐫. 𝐀𝐫𝐠𝐡𝐚 𝐒𝐞𝐧!

At the UbiNet: Ubiquitous Networked Systems Lab, IIT Kharagpur, we are proud to celebrate Argha, who has successfully defended his PhD thesis, "𝐀𝐝𝐯𝐚𝐧𝐜𝐢𝐧𝐠 𝐍𝐞𝐱𝐭-𝐆𝐞𝐧𝐞𝐫𝐚𝐭𝐢𝐨𝐧 𝐦𝐦𝐖𝐚𝐯𝐞 𝐒𝐞𝐧𝐬𝐢𝐧𝐠 𝐀𝐜𝐫𝐨𝐬𝐬 𝐀𝐮𝐭𝐨𝐦𝐨𝐭𝐢𝐯𝐞 𝐚𝐧𝐝 𝐇𝐮𝐦𝐚𝐧-𝐂𝐞𝐧𝐭𝐫𝐢𝐜 𝐄𝐧𝐯𝐢𝐫𝐨𝐧𝐦𝐞𝐧𝐭𝐬," under the supervision of 𝐏𝐫𝐨𝐟. 𝐒𝐚𝐧𝐝𝐢𝐩 𝐂𝐡𝐚𝐤𝐫𝐚𝐛𝐨𝐫𝐭𝐲.

Argha's work transforms low-cost millimetre-wave radar into a practical, privacy-preserving sensing modality, one that works without cameras or wearables, in the dark, through smoke, and across cluttered real-world spaces. Across his PhD, he built four complete systems spanning vehicles, homes, and industrial workplaces:

🚗 𝐦𝐦𝐃𝐫𝐢𝐯𝐞 (𝐈𝐄𝐄𝐄 𝐏𝐞𝐫𝐂𝐨𝐦 𝟐𝟎𝟐𝟑): passive, on-device detection of dangerous driving behaviours inside a moving vehicle.

🏠 𝐌𝐀𝐑𝐒 (𝐀𝐂𝐌/𝐈𝐄𝐄𝐄 𝐈𝐏𝐒𝐍 𝟐𝟎𝟐𝟒): room-scale, continuous tracking and activity recognition for multiple users using a single scanning radar.

🤖 𝐑𝐚𝐝𝐚𝐫𝐓𝐫𝐚𝐜𝐤 (𝐈𝐄𝐄𝐄 𝐒𝐦𝐚𝐫𝐭𝐂𝐨𝐦𝐩 𝟐𝟎𝟐𝟓): phase-based, radar-only ego-motion estimation that stays accurate even below the Doppler resolution limit, well suited to low-power robots and drones.

🦺 𝐌𝐈𝐑𝐎 (𝐀𝐂𝐌 𝐒𝐞𝐧𝐒𝐲𝐬 𝟐𝟎𝟐𝟔): multi-radar worker re-identification with personalised pollutant-exposure monitoring for occupational safety.

His impact reaches well beyond the thesis, with collaborative research on event-based eye tracking published at ACM IMWUT and NeurIPS, reflecting the breadth of his pervasive-computing expertise.

Congratulations, Argha, on a remarkable journey. We cannot wait to see what you sense next!

21/06/2026

Excited to share that our paper "𝐏𝐨𝐇𝐀𝐑: 𝐔𝐧𝐝𝐞𝐫𝐬𝐭𝐚𝐧𝐝𝐢𝐧𝐠 𝐇𝐲𝐩𝐞𝐫𝐥𝐨𝐜𝐚𝐥 𝐇𝐮𝐦𝐚𝐧 𝐀𝐜𝐭𝐢𝐯𝐢𝐭𝐢𝐞𝐬 𝐰𝐢𝐭𝐡 𝐏𝐨𝐥𝐥𝐮𝐭𝐢𝐨𝐧 𝐒𝐞𝐧𝐬𝐨𝐫 𝐍𝐞𝐭𝐰𝐨𝐫𝐤𝐬" has been accepted at IEEE DCOSS-IoT 2026 (the 22nd International Conference on Distributed Computing in Smart Systems and the Internet of Things), taking place this week in Reykjavik, Iceland (June 22 to 24).

What if the air-quality sensors already in our homes could tell us what people are doing, without any cameras, microphones, RF, or wearables? PoHAR repurposes commodity low-cost air-quality sensors as a privacy-preserving substrate for hyperlocal human activity recognition, running entirely on resource-constrained ESP32 microcontrollers.

The technical core is three contributions:

• Conflict-free data sharing: a replicated set (Set-CvRDT) that keeps self-supervised pollution embeddings consistent across nodes over lossy UDP, with guaranteed convergence even under node failure and message reordering.

• Pollution-aware on-device clustering: a hierarchical clustering algorithm on the ESP32 with a self-supervised distance metric that isolates only the sensors an activity actually affects, so concurrent activities in different zones stay separable.

• Leader-based hyperlocal inference: RAFT leader election with on-device inference using off-the-shelf ML, so the network recognises activities collaboratively with no central aggregation.

On real-world data, PoHAR reaches 97.41% accuracy for indoor activities and 99.68% for cooking, with inference under 34 microseconds on the ESP32. We validated it across 30 indoor sites over six months: 89.1 million samples, 13,646 hours, and 3,957 annotated activity events spanning occupancy, behaviour, and eleven cooking items.

Congratulations to Prasenjit Karmakar and Karthik Reddy Voddula, who led this work. Grateful to Prime Minister's Research Fellows (PMRF) and the Google Award on Society-Centered AI 2025 for their support.

Project page: https://ubinet-iitkgp.github.io/ubinet/pages/PoHAR.html

Code: https://github.com/prasenjit52282/PoHAR

Teaser video: https://youtu.be/COywxLYNXa8?si=nQmt8yfPAr1zEIsL

Low-cost air quality sensors are becoming ubiquitous in our daily l...

18/06/2026

🎉 Congratulations to 𝐃𝐫. 𝐓𝐢𝐫𝐭𝐡𝐚𝐧𝐤𝐚𝐫 𝐒𝐞𝐧𝐠𝐮𝐩𝐭𝐚 on successfully defending his PhD thesis at the Department of Computer Science and Engineering, IIT Kharagpur! 🎓

His thesis, "Auditability, Access Control and Redactability in Interoperable Blockchains," takes on some of the toughest trust challenges in cross-blockchain systems. The work delivers three connected contributions: making cross-chain transactions auditable and non-repudiable, enforcing end-to-end access control across permissioned consortiums, and enabling governed, verifiable redaction across interoperating chains.

Supervised by Prof. Shamik Sural and Prof. Sandip Chakraborty, his research has already appeared in leading IEEE venues, including IEEE Transactions on Services Computing, IEEE ICBC, and IEEE BRAINS.

We could not be prouder. 👏 Wishing you the very best for the journey ahead, Dr. Sengupta! 🚀

16/05/2026

🎉 Sayantan Kuila presented our paper "𝐆𝐫𝐚𝐝𝐢𝐞𝐧𝐭 𝐀𝐧𝐠𝐥𝐞𝐬 𝐟𝐨𝐫 𝐑𝐮𝐧𝐭𝐢𝐦𝐞 𝐃𝐞𝐭𝐞𝐜𝐭𝐢𝐨𝐧 𝐨𝐟 𝐁𝐨𝐝𝐲-𝐏𝐨𝐬𝐢𝐭𝐢𝐨𝐧 𝐒𝐡𝐢𝐟𝐭 𝐢𝐧 𝐇𝐮𝐦𝐚𝐧 𝐀𝐜𝐭𝐢𝐯𝐢𝐭𝐲 𝐑𝐞𝐜𝐨𝐠𝐧𝐢𝐭𝐢𝐨𝐧" at the FMSys Workshop @ IEEE/ACM CPS-IoT week 2026! 🚀

A wonderful collaboration with Soumyajit Chatterjee (Brave Software / University of Cambridge) and our team at the Department of Computer Science and Engineering, IIT Kharagpur.

📱 𝐓𝐡𝐞 𝐩𝐫𝐨𝐛𝐥𝐞𝐦: Wearable devices rarely stay in one place -- a smartphone might be in your pocket, on your arm, or in your hand. This body-position heterogeneity silently breaks on-device Human Activity Recognition (HAR), and the issue persists even with the new wave of wearable foundation models.

💡 𝐎𝐮𝐫 𝐚𝐧𝐠𝐥𝐞 (𝐥𝐢𝐭𝐞𝐫𝐚𝐥𝐥𝐲): Instead of asking "how do we adapt?", we ask "do we even need to adapt right now?" -- because continuous test-time adaptation is simply too expensive for resource-constrained wearables.

🔍 𝐖𝐡𝐚𝐭 𝐰𝐞 𝐟𝐨𝐮𝐧𝐝: ▪️ Body-position shifts cause sharp, activity-dependent performance drops (walking generalizes well across positions; sitting does not) ▪️ A lightweight gradient-angle analysis captures directional mismatch between reference and runtime gradients -- a far more meaningful signal than gradient magnitude alone ▪️ Last-layer gradients offer roughly an order-of-magnitude sharper separation between same-position and cross-position cases, while being memory-efficient enough for on-device monitoring

📈 𝐖𝐡𝐲 𝐢𝐭 𝐦𝐚𝐭𝐭𝐞𝐫𝐬: Gradient angles enable alignment-gated adaptation and angle-aware federated aggregation, moving beyond blind continuous adaptation toward smarter, runtime-aware policies for wearable AI.

Grateful to the FMSys Workshop organizers and the IEEE/ACM CPS-IoT week 2026 community for the engaging discussions! 🙏

🔗 https://fmsys-org.github.io/2026/index.html

13/05/2026

🎤 Sandip Chakraborty presented our paper "𝐌𝐈𝐑𝐎: 𝐌𝐮𝐥𝐭𝐢-𝐑𝐚𝐝𝐚𝐫 𝐈𝐝𝐞𝐧𝐭𝐢𝐭𝐲 𝐚𝐧𝐝 𝐑𝐚𝐧𝐠𝐢𝐧𝐠 𝐟𝐨𝐫 𝐎𝐜𝐜𝐮𝐩𝐚𝐭𝐢𝐨𝐧𝐚𝐥 𝐒𝐚𝐟𝐞𝐭𝐲" at 𝐀𝐂𝐌/𝐈𝐄𝐄𝐄 𝐒𝐞𝐧𝐒𝐲𝐬 𝟐𝟎𝟐𝟔! 🎉

Sharing a few moments from the presentation. 📸

Workers in hazardous environments like stone-cutting yards face serious exposure to particulate matter every day, yet monitoring pollution exposure at the individual level remains an open challenge, especially without compromising worker privacy through camera-based surveillance.

𝐌𝐈𝐑𝐎 tackles this with a privacy-preserving framework that pairs continuous PM sensing with a multi-radar mmWave re-identification backbone, tracking and re-associating workers across overlapping radar views -- no cameras involved. A GAN-based view adaptation model corrects azimuth-related distortions in range-Doppler signatures to keep identities consistent across radars, achieving a 90.4% re-ID F1-score, validated through field deployments in real industrial workspaces.

A big thank you to the wonderful SenSys community for the engaging discussions and thoughtful feedback during and after the talk.

Team behind MIRO: 👏 Tirthankar Halder (IIT Kharagpur) 👏 Argha Sen (IIT Kharagpur) 👏 Swadhin Pradhan (Cisco Systems) 👏 Rijurekha Sen (IIT Delhi) 👏 Sandip Chakraborty (IIT Kharagpur)

We gratefully acknowledge the support of the Department of Science and Technology (DST), Government of India, the National Science Foundation (NSF), and Google Research (Google's Society-Centered AI research funding) for making this work possible.

🎬 Teaser video: https://youtu.be/nTJj0PruMGk?si=Dcg_y5HLCcMeIJhg

🌐 Project page: https://ubinet-iitkgp.github.io/ubinet/pages/miro.html

🔗 Paper: https://dl.acm.org/doi/10.1145/3774906.3802761

23/04/2026

Our paper "𝐌𝐈𝐑𝐎: 𝐌𝐮𝐥𝐭𝐢-𝐫𝐚𝐝𝐚𝐫 𝐈𝐝𝐞𝐧𝐭𝐢𝐭𝐲 𝐚𝐧𝐝 𝐑𝐚𝐧𝐠𝐢𝐧𝐠 𝐟𝐨𝐫 𝐎𝐜𝐜𝐮𝐩𝐚𝐭𝐢𝐨𝐧𝐚𝐥 𝐒𝐚𝐟𝐞𝐭𝐲" has been accepted at 𝐀𝐂𝐌/𝐈𝐄𝐄𝐄 𝐒𝐞𝐧𝐒𝐲𝐬 𝟐𝟎𝟐𝟔!

Workers in hazardous environments such as stone-cutting yards face significant exposure to particulate matter, yet monitoring individual-level pollution exposure remains a challenge. MIRO addresses this through a privacy-preserving framework that combines continuous PM sensing with a multi-radar mmWave re-identification backbone -- tracking and re-associating workers across overlapping radar views without the use of cameras. A GAN-based view adaptation model corrects azimuth-related distortions in range-Doppler signatures to maintain identity consistency across radars, achieving a 90.4% re-ID F1-score validated through field deployments in real industrial workspaces.

Congratulations to the team: 𝐓𝐢𝐫𝐭𝐡𝐚𝐧𝐤𝐚𝐫 𝐇𝐚𝐥𝐝𝐞𝐫 (IIT Kharagpur), 𝐀𝐫𝐠𝐡𝐚 𝐒𝐞𝐧 (IIT Kharagpur), 𝐒𝐰𝐚𝐝𝐡𝐢𝐧 𝐏𝐫𝐚𝐝𝐡𝐚𝐧 (Cisco Systems), 𝐑𝐢𝐣𝐮𝐫𝐞𝐤𝐡𝐚 𝐒𝐞𝐧 (IIT Delhi), and 𝐒𝐚𝐧𝐝𝐢𝐩 𝐂𝐡𝐚𝐤𝐫𝐚𝐛𝐨𝐫𝐭𝐲 (IIT Kharagpur).

We gratefully acknowledge the support of the Department of Science and Technology (DST), Government of India, the National Science Foundation (NSF), and Google Research (Google' Society Centered AI research funding) for supporting this work.

🔗 Project page: https://ubinet-iitkgp.github.io/ubinet/pages/miro.html

🎬 Teaser video: https://youtu.be/nTJj0PruMGk?si=4jItmefizDbNErx9

Occupational exposure to airborne particulate matter (PM) is a major health risk in open industrial settings like stonecutting yards. Conventional solutions ...