$ ros2 launch keval portfolio.launch.py
[INFO] [launch]: Default logging verbosity is set to INFO
[INFO] [lidar_driver-1]: process started with pid [2019]
[INFO] [sensor_check-2]: process started with pid [2023]
[sensor_check-2] LiDAR ✓ Camera ✓ Radar ✓ GNSS ✓ IMU ✓
[fleet_manager-3] connected to 10+ AMRs · Temple, TX
[portfolio_node-4] loading experience: Meta · Tatum Robotics · The Innovation Story · KJSCE Robocon
[portfolio_node-4] loading projects: data-centre SLAM + Nav2 · sensor fusion · RL 7-DOF · +10
[portfolio_node-4] ready. welcome to kevalvisaria.com

click or press any key to skip

SYS.STATUS — FIELD OPERATIONS // ROS2

Keval Visaria

Keval Visaria

Robotics Field Engineer

I deploy, validate and operate autonomous mobile robots in high-availability production environments — currently at Meta (via Beyondsoft) in Temple, TX. Hardware, software, and the messy reality in between.

$ now: field-testing autonomous mobile robots▍

RVIZ2/map/scan/costmap/plan

Fleet in production

10+ AMRs

Sensors validated

LiDAR · Camera · Radar · GNSS · IMU

Core stack

ROS2 · C++ · Python

Day-to-day

Field testing · Root-cause analysis

01 / ABOUT

About me

I'm a robotics engineer who works where robots actually have to run: in production, outdoors, around people and machinery — not just in simulation. My current work is field robotics: system setup and validation, structured autonomy testing, sensor profiling, and debugging live robots with logs, ROS bags, and the tools that show what really happened.

That field work sits on an embedded foundation: firmware in C and C++, PCB design and bring-up, control systems, and sensor fusion. Whether it's a humanoid arm, a mobile robot, or a fleet of autonomous units, I enjoy bringing hardware, code, and mechanics together into systems that keep working after deployment.

Build it, test it in the field, fix what breaks, ship it again.

Field notes

ACTIVE

Role
Robotics Field Engineer
Where
Meta (via Beyondsoft) · Temple, TX
Focus
Field testing · sensor validation · fleet diagnostics
Foundation
Embedded C/C++ · PCB bring-up · sensor fusion
Building next
Autonomous exploration & patrol for my ROS 2 data-centre robot

Route so far

Mumbai2019 – 23B.Tech ElectronicsBoston2023 – 25MS RoboticsTemple, TX2025 –AMR fleet in production

02 / EXPERIENCE

Experience

November 2025 – PresentCURRENT

Robotics Field Engineer

Meta (via Beyondsoft)

  • Release-level field testing for a fleet of 10+ autonomous mobile robots
  • Sensor profiling & validation: LiDAR, cameras, radar, GPS/GNSS, IMU
  • Python & Bash automation for fleet diagnostics and log extraction
ROS2Field TestingSensor ValidationFleet OperationsPython+4

July 2024 – December 2024

Embedded Systems Engineer Intern

Tatum Robotics

  • Custom I2C drivers for haptic motor feedback on a humanoid arm
  • systemd services & timers that cut driver bring-up time
  • Circuit and PCB design for the arm's power management
Embedded SystemsCircuit DesignFirmware DevelopmentC/C++Linux+2

September 2022 – July 2023

Robotics Intern

The Innovation Story

  • 15% control-system efficiency gain using sensor fusion
  • Designed PCBs for 10+ robotics projects
  • Mentored high-school FTC & FRC robotics teams
RoboticsPCB DesignMotion ControlJavaMATLAB+2

August 2021 – August 2022

Electronics Team Lead

Team KJSCE Robocon

  • 95% pathfinding accuracy with IMU, encoder & IR sensing
  • 30% more efficient ZigBee/Bluetooth data link
  • Multi-microcontroller PCBs talking UART, I2C, SPI & CAN
PCB DesignSensor IntegrationMotion ControlCommunication ProtocolsRobotics+1

03 / COMPETITIONS

Competitions & mentoring

Where I learned to build robots under a deadline — and later helped students do the same.

MATCH FOOTAGE
COMPETITOR

2021 – 2022

ABU Robocon

Electronics Team Lead · Team KJSCE Robocon

Theme: Lagori (2022)

The Asia-Pacific Broadcasting Union Robot Contest: university teams design and build robots that complete high-speed, precision tasks against the clock — with autonomous and manually driven robots working together.

  • Led the electronics team: PCB design, sensor integration, motion control and communication
  • Multi-microcontroller PCBs talking UART, I2C, SPI and CAN for fast, synchronised motion
  • IMU, encoder and IR sensing for autonomous navigation — 95% pathfinding accuracy
  • ZigBee/Bluetooth link between robots — 30% more efficient data transfer
PCB DesignSensor IntegrationMotion ControlCANZigBeeTeam Leadership
MATCH FOOTAGE
MENTOR

2022 – 2023

FIRST Tech Challenge

Mentor · FTC team 22985 Paraducks · via The Innovation Story

Theme: POWERPLAY (2022–23)

FIRST's high-school robotics programme: student teams design, build and program robots to play a new game each season. I coached teams through FTC and FIRST Robotics Competition (FRC).

  • Mentored FTC team 22985 Paraducks through the 2022–23 POWERPLAY season
  • Coached high-school students through robot design, build and programming
  • Hands-on guidance on mechanisms, wiring and control code for autonomous and driver-controlled periods
  • Helped teams turn ideas into robots that could complete complex game tasks
MentoringSTEM OutreachJavaRobot DesignFTCFRC

04 / PROJECTS

Projects

Robotics work first — ROS 2 navigation, sensing and learning-based control — followed by hardware builds and computer vision.

01 · FEATUREDNEW · ROS 2

A custom differential-drive robot that maps data-centre halls with slam_toolbox and navigates them autonomously with Nav2 + AMCL in Gazebo — tuned for tight 1.2 m rack aisles, with keep-out zones around restricted areas.

SLAM map at 99.9% precision vs ground truth · ≤ 10 cm localisation error

ROS 2 HumbleNav2slam_toolboxAMCLGazebo
Code

02 · FEATURED

Estimated a car's heading, velocity and path from IMU, magnetometer and GPS data I logged with my own ROS drivers while driving around Boston — then dead-reckoned the route and compared it to GPS.

Hard/soft-iron calibration + complementary filter for heading

ROSSensor FusionIMUGPSComplementary Filter
Code

03 · FEATURED

Trained and compared PPO, SAC, TD3 and DDPG on a simulated 7-DOF KUKA arm reaching task in PyBullet, with a Dijkstra joint-space planner as a classical baseline.

SAC improved from about −191 to −20 average reward over 5,000 episodes

Reinforcement LearningPyBulletPPOSACTD3
Code

04 · FEATURED

Characterised an IMU's gyro and accelerometer noise from a 5-hour stationary recording using Allan variance, then compared the results with the VN-100 datasheet.

Separated white noise from flicker (bias-instability) noise on all three gyro axes

IMUAllan VarianceSensor CharacterisationMATLABPython
Code

05 · FEATURED

Calibrated a camera with a checkerboard, analysed reprojection error, and stitched photos of murals around campus into panoramas using Harris corners.

Full pipeline from intrinsic calibration to stitched panoramas

Camera CalibrationFeature DetectionHomographyMATLABComputer Vision
Code

More projects

HARDWARE

Designed and built a 3-DOF robotic arm with inverse kinematics for precision positioning and optimised motion control.

RoboticsInverse KinematicsServo Motors

HARDWARE

Designed a 4-legged robot with pneumatic actuation, a custom PCB and wireless control for navigating rough terrain.

RoboticsPneumaticsPCB Design

PERCEPTION

Real-time video processing in C++ with OpenCV: filters, face effects, selective colour and a faster separable blur.

C++OpenCVReal-time Processing
Code

PERCEPTION

Find visually similar images using colour histograms, texture filters and ResNet18 embeddings — and combinations of all three.

PythonPyTorchResNet
Code

PERCEPTION

Segment, describe and classify objects from a live camera feed, and teach the system new objects on the fly.

Computer VisionMachine LearningKNN
Code

PERCEPTION

Calibrate a camera, track its pose in real time and project 3D virtual objects onto a live video feed.

OpenCVCamera Calibration3D Projection
Code

PERCEPTION

A PyTorch CNN for MNIST digits, extended to Greek letters with transfer learning and run live on a webcam.

PyTorchCNNTransfer Learning
Code

PERCEPTION

Classical and deep-learning lane detection: Canny + Hough, sliding windows, curvature estimation and a U-Net trained on TuSimple.

OpenCVU-NetDeep Learning
Code

05 / EDUCATION

Education

Sep 2023 – Aug 2025

Northeastern University

Master in Robotics — ECE concentration

Boston, MA

Key coursework

Mobile RoboticsRobot Sensing and NavigationRobot Mechanics and ControlReinforcement LearningPattern Recognition and Computer Vision

Aug 2019 – Jul 2023

K.J. Somaiya College of Engineering

Bachelor of Technology in Electronics Engineering

Mumbai, India

Key coursework

Robotics and AutomationEmbedded System DesignControl System EngineeringMicroprocessors and PeripheralsDigital Signal Processing

06 / STACK

Skills

The stack I actually use in the field — not just the stack I studied.

Field & Autonomy

ROS2Field TestingSensor ValidationRelease TestingFleet OperationsRoot-Cause AnalysisRvizPlotJugglerROS Bag AnalysisPath PlanningMotion Control

Sensing & Perception

LiDARCamerasRadarGPS/GNSSIMUSensor FusionOpenCVPyTorchTensorFlowDeep LearningReinforcement LearningImage Processing

Embedded & Hardware

Embedded CC++STM32Jetson NanoRaspberry PiArduinoMicrocontrollersPCB Design4-Layer PCBPCBA Bring-upMotor ControllersCircuit Design

Software & Tools

PythonBashLinuxGitCI/CDDockerJTAGGDBJIRAMATLABJavaSOLIDWORKSAutoCAD3D Printing

07 / CONTACT

Let's talk robots

Always happy to talk field robotics, autonomy validation and embedded systems — or anything about robots that have to work in the real world.

LinkedIn github.com/d0stMalone Temple, TX · US Central time

© 2026 KEVAL VISARIA — FIELD ROBOTICS // TEMPLE, TX