Adapted and deployed the open-source STORM MPC framework on the 7-DOF xArm7 robot to enable real-time, joint-space motion planning and control. Implemented GPU-accelerated trajectory sampling and feedback-driven optimization at 100 Hz for reactive pose tracking and obstacle avoidance. Modified robot-specific kinematic models, control interfaces, and implemented filtering for robust state estimation and smooth motion generation.

Code

Developed a pose graph-based 2D Simultaneous Localization and Mapping (SLAM) system using nonlinear least squares optimization, integrating odometric and loop closure constraints to significantly improve robot trajectory accuracy. Utilized Python and libraries such as NumPy and g2o for algorithm implementation and performance evaluation against ground truth data. Code

Developed a dense 3D reconstruction pipeline using OpenCV's SGBM algorithm to compute disparity maps from stereo images, generating detailed colored point clouds with Open3D for autonomous robotic scene understanding. Code

Implemented advanced robotics planning algorithms for holonomic and non-holonomic robots in 2D simulated environments.

• Developed Rapidly-exploring Random Tree (RRT) algorithms for safe navigation around dynamic and static obstacles.
• Designed Model Predictive Control (MPC) using the MPPI update rule for unicycle kinematics under velocity, acceleration, and lane constraints.
• Modeled non-holonomic robot trajectories using Bernstein polynomials, incorporating obstacle avoidance and waypoint optimization with the CEM update rule.
• Generated simulation videos, trajectory visualizations, and performance plots to validate algorithms.

Code

Developed a precise camera–IMU calibration pipeline using ROS2 and Intel RealSense sensors to estimate extrinsic parameters, achieving low MSE in orientation estimation and enabling accurate sensor fusion for motion tracking. Code

Developed a novel iterative robust PCA model integrating SVD-based decomposition and Angular Differential Imaging (ADI) for separating faint planetary signals from stellar noise.

• Achieved uniform SNR enhancement and reduced quasi-static speckles on ESO–VLT SPHERE data of β Pictoris b.
Code

Tools: Python, NumPy, OpenCV, scikit-learn, Matplotlib

Explored classical and modern approaches for image understanding and visual perception.

• Image Classification: Implemented handwritten digit recognition on the MNIST dataset using both classical (SIFT + Bag of Visual Words + SVM) and deep learning methods (CNNs and Vision Transformers). Analyzed the effect of hyperparameter variations on accuracy and data efficiency.
• Detection & Tracking: Developed Viola–Jones face detection and IoU-based multi-frame tracking for a movie clip; implemented and fine-tuned YOLOv8 detectors on the Open Images dataset, comparing pretrained vs. scratch models and dataset sizes.
• Segmentation & Multimodal Learning: Trained and evaluated U-Net models on Cityscapes, investigating the impact of skip connections on mIoU. Compared CLIP (Contrastive Language–Image Pretraining) and ImageNet-pretrained ResNet-50 encoders; evaluated FP16 inference speedups and memory savings.

Tools: Python, PyTorch, OpenCV, scikit-learn, Ultralytics YOLO, Wandb, NumPy, Matplotlib

Code

Implemented classical and advanced image-processing algorithms from scratch and using OpenCV to develop a comprehensive understanding of spatial, frequency, and feature-space techniques.

• Core Image Operations: Implemented pixel-level brightness/contrast adjustments, grayscale ↔ pseudocolor conversion, and video manipulation without high-level library calls.
• Histogram & Quantization: Wrote custom histequalize(); performed bit-depth quantization; created encryption/decryption functions using LSB-based steganography.
• Filtering & Edge Detection: Developed mean, median, Gaussian, bilateral, Sobel, Prewitt, Roberts, and Laplacian filters manually; analyzed spatial vs. frequency-domain behavior.
• Morphological Processing: Segmented overlapping coins and verified signatures through custom erosion, dilation, and skeletonization routines.
• Segmentation & Feature Detection: Applied Otsu and adaptive thresholding, Hough line/circle detection, and Harris corner detection for geometric feature extraction.

Code Tools: Python, NumPy, OpenCV, Matplotlib, scikit-image

Developed multiple ML and AI pipelines from scratch and using libraries, demonstrating strong understanding of statistical learning theory, model evaluation, and neural architectures.

• Implemented Decision Trees, kNN, Naive Bayes, GMMs, Clustering Metrics (from scratch), and PCA Eigenface Recognition.
• Built CNN and Autoencoder models on CIFAR-10 and MNIST datasets, and RNN–LSTM networks for IMDB sentiment analysis.
• Final Kaggle mini-project on Age Prediction from Facial Images using transfer learning (VGG16), achieving strong leaderboard performance.

Code Tools: Python, NumPy, scikit-learn, PyTorch, Matplotlib, pandas

Developed an Android app that analyzes social media activity and psychometric test results to classify users as better suited for tech or sales roles. The app outputs a percentage match for each category, aiding recruiters in shortlisting candidates. Code

Built a movie recommender system using collaborative filtering and matrix factorization methods to provide personalized movie suggestions based on user ratings and preferences. Code

Implemented a classification model to categorize election-related social media posts, contributing to large-scale political discourse analysis under the Precog research group. Code

Worked on analyzing and classifying election-related advertisement data to estimate campaign expenditures, using text and image-based analysis pipelines. Code 1 Code 2

Designed a between‑subjects experiment investigating how immersive VR vs. real‑world environments affect spatial memory encoding and retrieval. Participants (ages 18–40) were trained and tested across three complexity levels (object placement, route navigation, maze navigation) in matched environments. Measured recall alignment error, recall time, and hypothesized brain activation patterns (hippocampus, MEC). Results demonstrated superior spatial recall in real‑world encoding, highlighting sensorimotor engagement's role in memory consolidation and VR's limitations in complex spatial tasks.

Report

Replicated and extended a Nature study on gender differences in cost-benefit decision-making by analyzing effort allocation across 81 participants using mixed-effects models, Kruskal-Wallis tests, and post-hoc analyses. The study confirmed that vagus nerve stimulation significantly enhances reward-driven motivation (p < 0.001) and demonstrated that gender independently modulates effort maintenance, providing new insights into the interplay between physiological interventions and behavioral decision-making.

Code Report