Robotics engineering graduate student with hands-on experience in ROS, SLAM, path planning, embedded systems, 3D computer vision (SfM, NeRF, VIO), and ML-driven perception—bridging simulation to real-world hardware.
I'm a Robotics Engineering M.S. student at Worcester Polytechnic Institute, with a strong foundation in computer science and hands-on experience building robotics and IoT systems. I've worked on ROS-based autonomy (SLAM, Gmapping, navigation), embedded sensing pipelines, and ML-driven safety applications— from simulation to hardware prototypes.
Fire detection + localization with robotic response: shortest path navigation, live video streaming, and mobile app control using Arduino Nano 33 BLE Sense + ESP32. Presented at ICIoT 2025.
Neural network trained on fire/non-fire datasets using color sensor patterns for early hazard prediction, demonstrating ML + IoT integration.
Implemented autonomous navigation of a skid-steer robot in Gazebo using SLAM, Gmapping, lidar + odometry, and ROS pub-sub architecture.
IoT system with ESP8266 + MQ135 sensor and cloud visualization (ThingSpeak) for real-time air quality monitoring.
Implemented a custom Rapidly-exploring Tree Planner (RTP) in C++ with collision checking for point + rotated square robots, tested on 10-link and 20-link manipulators.
Implemented planning for a chain-box robot in a 7-DOF state space (ℝ²×S¹×S⁴) with custom validity checks, tested RRTConnect and PRM* across challenging scenarios.
Compared kinodynamic planners on Pendulum and Car systems; implemented AO-RRT from scratch and benchmarked KPIECE1, SST, and AO-RRT.
Built a full TAMP stack: PDDL → Pyperplan → OMPL → Genesis simulator for a Franka Panda, with closed-loop re-planning and predicate extraction.
Built Pb-Lite from scratch with 112-filter banks, texton/brightness/color clustering, half-disk χ² gradients, and fusion with Sobel+Canny for robust boundaries.
Implemented CNN, ResNet, ResNeXt, DenseNet; best result: ResNet (92K params) with 84% test accuracy. Used AdamW + OneCycleLR over 50 epochs.
Implemented Zhang’s calibration from scratch: homographies, intrinsics, extrinsics (Rodrigues), nonlinear refinement with radial distortion.
Classical panorama stitching with cylindrical projection, ANMS, SSD matching + Lowe ratio, RANSAC homography, and distance transform blending.
Supervised + unsupervised HomographyNet with TensorDLT + STN warping; hybrid stitching combining classical features with DL homography estimation.
Reconstructed the 3D point cloud of a building from just 5 photographs — estimating camera positions, triangulating thousands of 3D points, and refining everything with bundle adjustment. A full SfM pipeline built from scratch.
Taught a neural network to understand 3D space from 2D photos. The model learns to render photorealistic novel views by predicting color and density at every point in a scene — achieving 27.4 dB PSNR on synthetic datasets and trained on custom real-world captures.
Converts raw dashcam video into a full 3D scene understanding: detecting vehicles, pedestrians, lanes, traffic signals, and road signs — then rendering everything in Blender with motion state estimation, brake light detection, and collision prediction.
Implemented both a classical Stereo Multi-State Constraint Kalman Filter (S-MSCKF) achieving 0.12m RMSE on EuRoC, and deep learning VIO networks (vision-only, IMU-only, fused) with gated sensor fusion — reducing trajectory error to 1.18m after global optimization.
Independently scoped, designed, and delivered digital solutions for small business clients end-to-end, managing the full project lifecycle with minimal supervision.
Contributed to robot behavior design and perception pipelines for the SoftBank Pepper Robot, integrating speech, gesture, and visual feedback modalities using ROS2.
Designed autonomous task pipelines and simulation environments in Webots for the national eYSRC competition, integrating perception and decision-making logic end-to-end.
Built ROS nodes and completed an end-to-end project designing differential drive and skid-steer robots with full autonomy capabilities.