Active perception · Contact-rich manipulation · Sim-to-real learning
I’m a PhD student at UC Davis with Prof. Stavros Vougioukas, building an active vision system that coordinates several cameras for multi-arm robotic harvesting.
I care about Physical AI that survives outside the lab: perception, multimodal fusion, and sim-to-real learning that hold up in sunlight and occlusion. I’m also building a startup on a simulation and synthetic-data platform for orchard robots.
If you’re building or researching robots for the real world, let’s talk.
July '26 — Attended Y Combinator Startup School 2026, featuring talks from Jensen Huang, Sam Altman, Alexandr Wang, and Jeff Dean. Had a great time meeting and connecting with builders from across the startup community.
July '26 — Started PhD at UC Davis under Dr. Stavros Vougioukas. I will be working on developing active-vision for multi-arm apple harvesting.
Jan '26 — I collaborated with Parv Maheshwari and Dr. Wenshan Wang from AirLab (CMU) for the Anythermal work (anythermal.github.io). Our paper got accepted into the International Conference on Robotics and Automation (ICRA) 2026.
Dec '25 — Attended my first IIT Kharagpur Alumni Meet at Dr. Deepak Khazanchi's home in Omaha, USA, and had the opportunity to meet our director, Dr. Suman Chakraborty.
July '25 — I attended ASABE 2025 in Toronto. I was a co-instructor of our team's CPD workshop on "Developing Custom Autonomous UGVs and UAVs from the Ground Up for Hands-on Learning". Also, I presented my work on "Virtual Farm Environments and Sim-to-Real Transfer in Agricultural Robotics Using NVIDIA Omniverse".
June '25 — I attended CVPR 2025 in Nashville. One of the best experiences till now.
Dec '24 — Officially graduated from IIT Kharagpur with a B.Tech in Agricultural and Food Eng., Micro-Specialization in Artificial Intelligence, and M.Tech in Financial Eng.
May '24 — I was awarded the Institute Order of Merit by Technology Students' Gymkhana IIT Kharagpur based on my participation in InterIIT events, competitions, and research papers.
May '23 — Worked in the summer at AI2Lab, University of Calgary working on developing a Garbage Classification System using Image and Text!
Mar '23 — Selected as a Mitacs Research Intern at the Department of Electrical and Software Engineering, University of Calgary.
Feb '23 — Participated in the IIT Kharagpur Contingent at the Inter IIT Tech Meet 11.0.
Jul '22 — Paper got selected for the Fourth International Workshop on Machine Learning for Cyber-Agricultural Systems (MLCAS2022).
Jul '22 — Honoured to receive the prestigious Vinod Gupta Leadership Scholarship Award.
Jul '22 — The work of last summer with Purdue University got selected for the American Society of Agricultural and Biological Engineers Annual International Meet oral presentation (ASABE2022).
May '22 — The work for this summer got selected for MLRC 2022.
May '22 — Spending the summer at SCS Lab, Iowa State University working on Self-Supervised Multimodal Data Fusion Strategies!
Mar '22 — Won 1st place at the Inter IIT Tech Meet 2022 at the event 'Silicon Labs' High-Prep Social Entrepreneurship Challenge'.
Jan '22 — Honoured to receive the prestigious Honda Young Engineer and Scientist's (Y-E-S) Award organized by Honda Foundation, Japan.
Building an active-vision system for multi-arm fruit harvesting: three arm-mounted cameras learn where to look next to find apples hidden behind leaves. Sim first, then the orchard.
New: DLCL, our UAV–UGV cooperative localization work, is accepted in the International Journal of Computer Vision, and was presented at the IROS 2026 LTP Workshop.
Open to collaborations on perception for robots in unstructured environments. Say hello.
Developing an active-vision system for multi-arm fruit harvesting. A ground vehicle approaches a tree; three cameras on independent arms choose the next best view, fuse detections into one shared map of the canopy, and reveal fruit hidden behind leaves and branches—built and tested in simulation first, then transferred to real orchards.
Worked on the project "AnyThermal: Universal Thermal Perception Backbone" and contributed to developing a robust RGB–thermal alignment pipeline using stereo depth back-projection from the FoundationStereo model, enabling dense pixel-level registration across 16,943 synchronized RGB–T pairs.
GRA in the MAARS Lab at UNL, focusing on designing a zero-shot sim-to-real reinforcement learning agent for robotic crop interaction and manipulation.
Also collaborated on building and deploying an MPC-controlled UAV platform, achieving 8–20 cm RMS tracking error on field trajectories. Implemented deep learning–based cooperative localization for UAV–UGV teams, improving position accuracy by 1 m over GNSS.
Worked under the supervision of Prof. Soumik Sarkar and investigated self-supervised multi-modal data fusion strategies using the MultiMAE model on crop canopy reflectance data (RGB, thermal, and depth images) for calculating the wilting score of plants.
Worked on another project applying reinforcement learning for robotic manipulation based on the finite element analysis model of a maize plant.
Volunteered in the collection of RGB, thermal, multispectral imagery, and ASD data using an RGB drone (Zenmuse X5S) and a hyperspectral spectroradiometer at Muscatine Island Research Farm.
Worked under the supervision of Prof. Peeyush Soni and implemented YOLO-v7 to detect apples from RGB images, achieving a mAP of 90.5%.
Implemented a modified A* algorithm for path planning of an ag bot moving in a field with trees at random locations.
Implemented inverse kinematics in Python for robotic manipulation in a Webots virtual environment.
May '21 – July '21
Worked under the supervision of Prof. Dharmendra Saraswat on determining the generalizability of object detection networks.
Explored methods for training and generalizing object detection networks using RGB images collected from fields with complex real-life differences.
Examined data augmentation techniques to increase the dataset and improve performance. The final model is robust enough to map or calculate the yield of soybean plants in various real-life situations.
May '21 – July '21
Worked under the supervision of Prof. Lav Khot on assessment of the percentage of failed crop germination: formulating non-germination at high throughput so it can guide replanting.
Explored variants of YOLOv4 to find the missing index of mulch planting, the percentage of mulch plants per unit area that will not lead to productivity.
Implemented active learning to reduce the burden of manually annotating imagery datasets.
Estimated the maximum resolution of testing images and used ground sampling distance (GSD) to calculate a range of drone flight heights that reduce inference time.
Joined as a mechatronics team member and created a chassis of a 6-wheeled planetary rover in SolidWorks for the University Rover Challenge (URC).
Led the DL Trajectory Prediction team with a focus on multi-modal trajectory prediction. Participated in the Machine Learning Reproducibility Challenge; the work was selected in MLRC 2021–22.
Jan '25 – May '25
AGEN 892: Applied Robotics in Agriculture and Natural Resources
Facilitated weekly hands-on lab sessions on ground and aerial robotics, guiding students through ROS-based concepts.
Appointed Student Representative of the IEEE Robotics and Automation Society Technical Committee on Agricultural Robotics and Automation (AgRA) for 2026–2027.
Represent student members in TC activities, help connect early-career researchers with the AgRA community, and support outreach around agricultural robotics and automation.
GYWS is a registered voluntary non-governmental organization run by students of IIT Kharagpur which works for the socio-economic welfare of underprivileged sections of society. The society runs Jagriti Vidya Mandir (JVM) near the village of Gopali, providing free education to 230 children.
I was part of the Sponsorship team, which plans and executes fundraising events. I coordinated a team of ten juniors, contacted companies for CSR, reached alumni of premier institutes, and introduced gift-matching programs. The effort raises nearly 45 lakhs per annum to cover the educational expenses of the students.
Elected UG Department Representative for the Agricultural and Food Engineering Department, IIT Kharagpur, for the academic session 2021–22.
Member of the Undergraduate Council, which acts as a liaison between students and the administration on academics and student welfare.
Mentored juniors working on applications of AI in agriculture and connected them with alumni and seniors.
GRASP replicates the agronomist's maize push test through a hierarchical perception–interaction–inference pipeline. A YOLO segmentation model, trained on SAM-3-generated labels from real field imagery, identifies the stalk; a classical IK-based controller then guides the manipulator through approach and alignment. A recurrent neural network, trained on interaction data collected in a GPU-accelerated deformable-body simulation with randomized FEM plant mechanics, infers soft/hard stalk stiffness from proprioception alone without requiring vision at inference. The system transfers to a real xArm6 + Bunker Pro robot setup, producing stiffness estimates that align with observed stalk lignification and maturity trends.
NeuroFly is an autonomous UAV navigation framework that combines neuromorphic computing and artificial intelligence to enable ultra-efficient real-time drone control on embedded hardware. By training a neural network to imitate an expert model predictive controller and converting it into a spiking neural network deployable on BrainChip Akida neuromorphic hardware, NeuroFly achieves accurate trajectory tracking with lower computational cost, latency, and energy use. The system was validated in simulation and real flight, with up to 97× faster inference than a conventional neural network while maintaining stable autonomous flight on a resource-constrained UAV.
A cooperative localization framework for heterogeneous UAV–UGV teams operating in GNSS-challenged environments. DLCL integrates deep learning-based multi-robot detection, an adaptive perception recovery mechanism, and GNSS–vision fusion. The work introduces a dataset of 30k UAV–UGV images and a 7.5k-frame sequential dataset for temporal modeling. A recurrent LSTM-based perception recovery module predicts future confidence trends and dynamically adjusts detection thresholds. Experiments with YOLOv8/v9/v10/v11 show perception recovery improving detection accuracy by up to 50% across diverse trajectories, particularly under low illumination and small-object conditions. A GNSS-prior cropping strategy reduces compute from 4.04 to 1.85 GFLOPs and inference latency from 36.74 to 21.86 ms. Adding cooperating agents monotonically reduces global RMSE and tightens the 3σ uncertainty bounds.
AnyThermal is a thermal backbone that captures task-agnostic thermal features for cross-modal place recognition, thermal segmentation, and monocular depth estimation. Feature representations from visual foundation models such as DINOv2 are distilled into a thermal encoder using thermal data from multiple environments. To bridge the diversity gap of existing RGB–thermal datasets, the work introduces TartanRGBT, an open-source data collection platform with synced RGB–thermal image acquisition.
A custom uncrewed aerial vehicle for precision agriculture, built for modularity, adaptability, and affordability. The platform integrates a Cube Blue flight controller for low-level control with a Raspberry Pi 4 companion computer that runs model predictive control for high-level trajectory optimization. Kalman filtering enables adaptive mission planning and real-time coordination with a moving uncrewed ground vehicle. Root mean square error stayed between 8 and 20 centimeters during autonomous operations, and the UAV followed the UGV along nonlinear, curved paths.
A cooperative localization framework for multi-robot systems of UAVs and unmanned ground vehicles. The method uses YOLOv8 for real-time detection and combines perception with Kalman filtering to improve localization accuracy in challenging environments.
Undertaken as part of the Machine Learning Reproducibility Challenge 2021, this review of an accepted ICCV 2021 paper checks reproducibility and empirical claims. Contributions include new sampling methods, new visualizations, results that improve on the original paper, and a look at how the model generalizes to other datasets.
An agent was trained to bend a corn plant built in the Chrono simulator, which computes finite element analysis. This project was done at Iowa State University.
Explored two methods for generalizing YOLOv4 using images from different fields, including confusing real-world differences such as weeds.
Implemented active learning to reduce the burden of manually annotating the imagery dataset by 75%.
Implemented a reactive gap-following algorithm on a physical F1Tenth autonomous vehicle. The method processes LiDAR data to identify the closest obstacle, creates a safety bubble around it by zeroing points within a radius, and then detects the largest gap free of obstacles. A rolling average filter reduces sensor noise, and a disparity function accounts for the vehicle's physical dimensions so the navigable gap is effectively widened.
The vehicle adjusts its speed from the steering angle to stay stable while navigating a track with obstacles. After simulation testing, the algorithm was deployed on the real car. Bubble radius, speed, and LiDAR range were tuned to prevent oscillations, and the vehicle completed three laps around the track without a global map.
A vision-guided control system for the uArm Swift Pro, built in ROS and Python. An Intel RealSense camera detects eggs and estimates their positions. Inverse kinematics turns those targets into joint angles so the arm moves smoothly. The vacuum gripper picks up and releases eggs, and ROS topics keep perception, motion planning, and actuation in sync for pick-and-place in an agricultural setting.
A three-day program on robot operation: programming, operating, troubleshooting, and optimizing collaborative and industrial robots.
An autonomous egg detection and collection system on ROS 2. LiDAR supports path planning, and a camera runs YOLOv11s to find and track eggs so they stay centered in the frame. An Arduino receives data from the LiDAR and camera nodes and drives the machine with PID controllers.
A robot with two motors and three ultrasonic sensors (front, right, left) for obstacle avoidance and wall following. A PID controller adjusts motor speeds so the robot holds a target distance from walls and stops when something is in front. Done in AGEN 836 under Prof. Santosh Pitla.
A modified A* algorithm for an ag bot moving through a field with trees, plus an ikpy controller for robotic manipulation in Webots. Bachelor's thesis work.
A modified A* algorithm for an ag bot moving through a field with trees, plus an ikpy controller for robotic manipulation in Webots. Bachelor's thesis work.
Undertaken as part of the Machine Learning Reproducibility Challenge 2020, a review of an accepted ECCV 2020 paper. Computational experiments reproduced the reported results within 5% of the claimed numbers.
Worked on the Mars rover team of the Autonomous Ground Vehicle Research Group, IIT Kharagpur.
Developed the wheel, chassis, and suspension for a rover prototype with 15° gradeability and a max speed of 20 cm/s.
Designed a 5-DOF modular robotic manipulator with a 2-finger grip for semi-autonomous on-board equipment repair.
Selected for the David H. and Annie E. Larrick Student Travel Award, granted on a competitive basis to fund graduate students who are personally presenting their research or scholarly activity at professional meetings.
Given by Technology Students' Gymkhana, IIT Kharagpur, based on involvement in Inter IIT events and research papers.
This scholarship is awarded every year to 14 students from the top 5 IITs, for excellence in science and technology.
Awarded the Vinod Gupta Leadership Scholarship for the academic session 2021–2022 based on leadership in the department, halls of residence, and campus activities during the first and second years.
Co-organized by IIT Kharagpur and Silicon Labs, India, as part of Inter IIT Tech Meet 10.0. The challenge was to develop an IoT-based truck health monitoring system.
Our team won first place and the Inter IIT Tech Meet 10.0 overall General Championship.
Won gold at Inter IIT Tech Meet 9.0, Drishtee’s tech-led Innovation for Rural Entrepreneurs. The team developed a platform connecting self-help groups to bulk buyers and building a network of rural producers.
DLCL: Deep Learning-enabled Cooperative Localization for UAV-UGV TeamInternational Journal of Computer VisionDetails
Custom UAV with Model Predictive Control for Trajectory TrackingFrontiers in Robotics and AIDetails
Digital soil mapping of available phosphorusSmart Agricultural TechnologyDetails
Detection and Occlusion-based Classification of Kashmiri Orchard ApplesJournal of Biosystems EngineeringDetails
From Goals, Waypoints & Paths To Long Term Human Trajectory ForecastingReScience CDetails
AnyThermal: Towards Learning Universal Representations for Thermal PerceptionICRA 2026Details
NeuroFly: Onboard Neuromorphic Spiking Neural Network Control for Autonomous UAV NavigationInternational Conference on Neuromorphic Systems · under reviewDetails
Cooperative Localization of UAVs Using Deep Learning-Based DetectionAIAA SCITECH 2025Details
Virtual Farm Environments and Sim-to-Real Transfer Using NVIDIA OmniverseASABE 2025Details
Fine-tuning YOLOv4 for Soybean detection in UAS imagesASABE 2022Details
Robotic Manipulation for Plant Interaction Using Reinforcement LearningMLCAS 2022Details
Autonomous Navigation on an F1Tenth PlatformReactive gap followingDetails
Vision-Guided Egg Manipulation with uArm Swift ProROSDetails
Universal Robots CORE TrainingCollaborative robotsDetails
EggBotAutonomous egg detection and retrievalDetails
Ultrasonic Sensor based NavigationAGEN 836Details
Path PlanningModified A* for an ag botDetails
Autonomous Apple HarvestingBachelor's thesisDetails
Endpoint Conditioned Trajectory PredictionMLRC 2020Details
University Rover ChallengeAGV, IIT KharagpurDetails
Larrick Graduate Student Travel AwardUniversity of Nebraska–LincolnDetails
Institute Order of MeritIIT KharagpurDetails
Honda Y-E-S AwardHonda Foundation, JapanDetails
Vinod Gupta Leadership ScholarshipIIT Kharagpur · 2021–22Details
Silicon Labs Social Entrepreneurship ChallengeInter IIT Tech Meet 10.0 · 1st placeDetails
Drishtee Innovation for Rural EntrepreneursInter IIT Tech Meet 9.0 · goldDetails
