Developer Portfolio

Experience

VIO Using MSCKF

I built a stereo vision-aided odometry system using a Multi-State Constraint Kalman Filter (MSCKF) that fuses IMU and camera data for precise real-time trajectory estimation, achieving a ~0.07 m median error on the EuRoC dataset.

Deep VIO

I created a deep learning-based VIO framework combining CNNs, LSTM, and FlowNet to fuse visual and inertial data, reaching ~79% accuracy for relative pose prediction on the validation set.

Edge Detection

I developed a Probability of Boundary method that outperforms classic edge detectors by leveraging texture, brightness, and color cues, and also compared four CNN architectures on CIFAR-10 under data augmentation and hyperparameter tuning.

Image Stitching

I developed a panoramic image stitching pipeline, combining classical methods (corner detection, homography via RANSAC) and deep-learning-based (supervised and unsupervised) approaches to seamlessly warp and blend multiple images into a single panorama.

Sfm & NeRF

I implemented a Structure-from-Motion workflow—estimating camera poses, 3D points, and bundle adjustment—and integrated a Neural Radiance Field (NeRF) for volumetric rendering, enabling novel view synthesis from sparse image inputs.

Camera Calibration

Using Zhang's Camera Calibration, I estimated intrinsic and extrinsic parameters (including radial distortion), then minimized reprojection errors to achieve precise camera calibration and consistent image rectification.

Vision-Based Robotic Manipulation

Designed a robotic grasping system for conveyor environments using YOLO for object detection and GGCNN for grasp prediction, achieving 92% accuracy. Implemented closed-loop pose correction to enhance grasp stability under conveyor vibrations.

Education

Master's of Science

Robotics Engineering

Worcester Polytechnic Institute

2023 - 2025

Bachelor's of Technology

Mechanical Engineering

NMIMS, India

2018 - 2022

Anuj Jagetia

Roboticist Specialized in Machine Learning