HELLO_WORLD β€” I BUILD ROBOTS THAT THINK

Jay Prajapati

Robotics Software Engineer

I turn caffeine into software modules and fiber into robots β€” designing autonomy stacks across motion planning, SLAM, perception, and controls with ROS/ROS2, C++, and Python.

M.Eng.
Robotics, UMD College Park
2+
Patents & publications
14+
Robotics projects shipped
scroll_to_explore
// 01

About Me

I'm a robotics engineer who lives at the intersection of software and hardware β€” equally at home writing a real-time C++ motion planner and torquing the last bolt on a robot I designed in CAD. I hold a Master of Engineering in Robotics from the University of Maryland, College Park and a B.Tech in Mechanical Engineering, a combination that lets me own the full stack: mechanism design, embedded control, and the autonomy software on top.

My work spans legged locomotion (a state-championship-winning, patented quadruped robot), swarm robotics (a 25-robot TurtleBot3 formation system in ROS2 with 99% geometric accuracy), SLAM and navigation (mapping 20,000 sqft environments, EKF sensor fusion, GPS/SLAM hybrid rovers), and deep-learning perception (YOLO-based tracking at 99 ms inference, semantic segmentation, real-time image captioning for accessibility).

As a Graduate Teaching Assistant at UMD, I mentored 95+ graduate students in computer vision and software engineering for robotics β€” homography, particle filters, sensor fusion, modern C++, CI/CD, and software design patterns. I care deeply about writing robot software that is tested, documented, and deployable, not just demo-ready.

Jay Prajapati
node/jay_prajapati
locationUnited States
educationM.Eng Robotics, UMD
focusAutonomy Β· Planning Β· Perception
status● all systems nominal
// 02

Experience & Education

$ history --experience

Aug 2023 β€” May 2024
Graduate Teaching Assistant
University of Maryland, College Park
  • Mentored 95+ graduate students in computer vision: homography, calibration, stereo vision, optical flow, Kalman filters, particle filters, sensor fusion, and SLAM.
  • Instructed advanced C++, OOP, ROS1–ROS2 bridging, unit testing, CI/CD, version control, and software design patterns with Git/GitHub workflows.
Jan 2020 β€” Jan 2022
Robotics Intern
Robotics Lab, BVM Engineering College β€” Anand, India
  • Pioneered a patented quadruped robot: canine-inspired gaits (walk, trot, turn, slope climbing), 40% weight reduction via 3D-printed structure, 90% gait accuracy vs. simulation.
  • Optimized multithreaded control of 12 servos on ATmega2560 with PCA9685 over I2C.
  • Built a LiDAR + camera wheeled robot on Raspberry Pi mapping up to 20,000 sqft with autonomous GPS navigation; benchmarked HectorSLAM, Cartographer, and Gmapping (Cartographer: 30% fewer artifacts).
  • 3D-scanned an 11 ft statue to a 98% accurate CAD model, cutting production costs 50% for FDM-printed miniatures.

$ history --education

Aug 2022 β€” May 2024
M.Eng, Robotics
University of Maryland, College Park β€” GPA 3.7/4.0
  • Robot design, perception, planning, machine learning, deep learning, software development.
Aug 2018 β€” May 2022
B.Tech, Mechanical Engineering
Gujarat Technological University, India β€” GPA 7.88/10
  • Industrial automation, robot design, rapid manufacturing and prototyping.
// 03

Projects

From legged locomotion to swarm coordination β€” hardware built, software shipped, metrics measured.

Multi-robot swarm system

Multi-Robot Swarm System

Swarm simulation of 25 TurtleBot3 robots in Gazebo/ROS2 achieving geometric formations with 99% accuracy via a master–slave coordination algorithm. CI/CD pipelines at 89% code coverage.

ROS2GazeboC++CI/CD
Path planning visualization

Dynamic Path Planning with RRT*

A*, Dijkstra, D*, RRT, and RRT* planners for TurtleBot3 in Gazebo, with probabilistic inference to predict dynamic obstacles β€” Bayesian modeling improved prediction robustness by 20%.

C++PythonRRT*Bayesian Inference
Real-time image captioning

AI Image Captioning for Accessibility

Deep-learning captioning for visually impaired users β€” Inception V3 encoders at 92% accuracy on real-time video, optimized with Block Static Expansion and multi-headed attention, with voice narration on mobile.

PyTorchCNNAttentionAccessibility
Semantic segmentation of food items

Real-Time Food Segmentation

RNN-based ResNet50 + PointRend achieving 80% segmentation accuracy on live video; transfer learning reached 67% test accuracy on a 4,935-image dataset, beating a custom model's 43%.

ResNet50PointRendTransfer Learning
[ yolo_v3 Β· kcf_tracker ]

Human Tracking Module

C++ real-time human detection and tracking: YOLOv3 at 84% confidence / 99 ms inference on MS-COCO, KCF tracking across 164k images, with Travis CI + Coveralls pipelines feeding a multi-robot planner.

C++YOLOv3KCFCI/CD
Autonomous pick and place robot

Autonomous Pick & Place Robot

Pi Camera color detection + A* planning with MPU6050 IMU, encoders, PID control, and ultrasonic obstacle avoidance β€” 80% goal-placement success in dynamic environments.

A*PIDOpenCVEmbedded
[ slam ⇄ gps Β· ekf_fusion ]

Indoor/Outdoor Autonomous Rover

Rover switching seamlessly between Gmapping SLAM (LiDAR + odometry) indoors and GPS waypoint navigation outdoors, with EKF sensor fusion minimizing drift and custom-fabricated aluminum mounts.

SLAMGPSEKFROS
[ homography Β· vanishing_point ]

TurtleBot3 Perception Stack

Monocular camera calibration, floor homography, and horizon-line estimation for waypoint following on a TurtleBot3 WafflePi, plus YOLO-based stop-sign detection across varying orientations.

Camera CalibrationHomographyYOLO
2D mapping comparison

2D SLAM Algorithm Benchmark

Comparative study of HectorSLAM, Cartographer, and Gmapping across a 20,000 sqft floor β€” Cartographer proved most accurate with 30% fewer map artifacts.

SLAMLiDARBenchmarking
[ moveit2 Β· rviz2 Β· dds ]

FRCobot ROS2 Bring-Up

ROS2 control stack for a collaborative arm: custom Python nodes for joint control and safety, URDF kinematics in RViz2, MoveIt2 motion planning for pick-and-place, validated in Gazebo before hardware.

ROS2MoveIt2URDFDDS
[ flask Β· docker Β· openrave ]

Robot Fleet REST API

Dockerized Flask API managing OpenRAVE robot configurations β€” full CRUD over JSON robot specs with parameter validation and consistent containerized deployment.

PythonFlaskDockerREST
IoT wastewater detection system

IoT Wastewater Spillage Detection

Co-invented a patented IoT system detecting abnormal spillage patterns via distributed sensors, streaming real-time alerts to a centralized monitoring platform. Published in IOP Journal of Physics.

IoTSensorsPatented
3D printed miniature statues

3D-Scanned Miniature Statues

Led 3D scanning of an 11 ft statue to a 98% accurate CAD model via Fusion 360 reverse engineering, then optimized FDM printing to produce 1000+ miniatures at 50% lower production cost.

3D ScanningFusion 360FDM Printing
// 04

Tech Stack

⌨ Languages

PythonC/C++MATLAB SQLBashMulti-Threading Data Structures & AlgorithmsUnit Testing PLC Programming

πŸ€– Robotics

ROS / ROS2GazeboRViz MoveItNav2SLAM Motion PlanningSensor Fusion (EKF) URDFControls / PID

🧠 AI & Perception

PyTorchTensorFlowKeras OpenCVCUDAscikit-learn YOLOSemantic Segmentation Camera CalibrationKalman / Particle Filters

βš™ Tools & Hardware

LinuxGitDocker CI/CDFlaskEigen SolidWorksPTC CreoFusion 360 3D PrintingEmbedded (AVR / RPi) DoxygenLaTeX
// 05

Patents & Publications

πŸ“œ

Quadruped Canine Robot β€” Patent

Application β„–202221001158 Β· The Patent Office Journal No. 11-2022, March 2022.

πŸ“„

IoT-Based Wastewater Spillage Detection System

IOP Conference Series: Journal of Physics β€” co-inventor of the patented system. read paper β†—

// 06

$ ping jay

Whether it's autonomous systems, motion planning, perception β€” or a robot that needs both a mechanical fix and a software one β€” my inbox is open. Let's build the future, one robot at a time.