Building intelligent systems that touch the physical world.

I’m a Robotics/AI engineer working across perception → planning → control. I design full‑stack systems that fuse sensors, classical controls, and modern ML to deliver robust behavior in real environments.

Intro Video

ROS 2 / MoveIt 2 C++ • Python Perception • Control GPU / CUDA
Soutrik headshot
(Topology Optimized Picture of my Face)

Timeline

Skip to projects →

  1. 2018 – 2022 | NIT Rourkela (India)

    Industrial & Product Design (Minor in Mech. Engg.)

    Graduated with the Gold Medal of Academic Excellence.

    Publications & Projects

    • Design → FEA/CFD → Deep-Learning pipeline for defect prediction.
    • Robotics path planning with ROS 2, Gazebo, and CV (supervised learning).
    • Rotating cylindrical fish-descaler for agro-forestry; 12 custom designs for different scale types; adopted by 2 wholesale markets.

    Internships

    • Indian Academy of Sciences — Summer Research Fellowship.
    • IIT Kanpur — Researcher on a major project.
    • Indian Space Research Organization — Embedded Systems Intern.
    CAD → FEA/CFD ROS 2 Computer Vision DL Pipelines

  2. 2022 – 2024 | University of Pennsylvania (Mechatronics)

    Admits: Penn, Columbia, Duke, Dartmouth, others

    Chose UPenn for an MSE in Mechanical Engineering (Mechatronics).

    Publication

    • Structural Topology Optimization of Headphones — treats “designs as data points in 3D space”; optimizes stiffness-to-material-cost; predictive analysis via deep learning.

    Experience

    • Researcher, Modular Robotics Lab — GRASP Lab (robotic gripping, motion planning, computer vision).
    Topology Opt. Optimization Deep Learning GRASP Lab

  3. 2024 – Now | Harrisburg University (Computer Science, AI)

    Expected Graduation: Feb 2026

    Transferred with UPenn approval to an MS in CS (AI focus). Building end-to-end intelligent robot pipelines.

    Master’s Thesis

    • DeepSeek-R1 LLM in ROS 2 Manipulation (advisor: Prof. Mani Akella) — command parsing, context management, and execution in MoveIt 2/Gazebo.

    Media

    • YouTube channel documenting the thesis (intro on home page; three ROS 2 demo videos in the project card).
    ROS 2 LLM MoveIt 2 Gazebo

Featured Projects

Work with me →
LLM to ROS 2 manipulation pipeline screenshot
ROS 2 MoveIt 2 LLM

Master's Thesis: Natural-Language Robot Manipulation

Built an LLM→ROS 2 bridge with command parser, context manager, and trajectory execution in Gazebo/MoveIt 2. Supports multi-user control with tests and CI.

Code Demo Project Brief (PDF)

Demo Videos

Bi‑Directional Drone
Gazebo Path Planning Autonomy

Publication DOI: 10.35940/ijrte.B6393.0910321 - Bi‑Directional Drone: Design + Path Planning

Concept + simulation of a drone that can carry payload above and below the frame with autonomous navigation.

Paper (PDF)
Fish Descaler
CAD FEA Embedded

Publication DOI:10.35940/ijrte.B6395.0910321 - Rotating Cylinder Fish‑Descaler

Affordable, mass‑manufacturable descaling system — perforated stainless drum, strap drive, and wash cycle.

Paper (PDF)
Topology Optimized Headphones
Topology Opt. FEA Additive Mfg

Publication DOI: DOI:10.35940/ijrte.F7462.0712223 - Structural Topology Optimization of Headphones

Iterative stiffness‑to‑mass optimization with OptiStruct/Altair — design space reduction and manufacturable forms.

Paper (PDF)

Work Experience

Full resume →

  1. Graduate Researcher | GRASP Lab, University of Pennsylvania

    March 2023 – March 2024 · Philadelphia, PA

    Perception → planning → control for a Fetch mobile manipulator; reliability and autonomy improvements.

    • Developed the perception‑planning‑control loop for a 7‑DOF arm on a Freight base (11 total DOF) to operate unsupervised indoors.
    • Fused LiDAR + RGB‑D + IMU for navigation, tuned motion‑planning to avoid stalls, hardened ROS 2 messaging over flaky Wi‑Fi, and built a regression harness to prove improvements.
    • Integrated SMORES‑EP modular faces for contact‑aware grasps and re‑tries on odd shapes; onboard processor ran Kalman filters with centralized Wi‑Fi control.
    System setup & environment — GRASP Lab (ROS 2, sensors & bring-up).
    Perception → planning → control loop on Fetch (ROS 2).

  2. Embedded Systems Intern | ISRO — Indian Space Research Organisation

    Sep 2021 – Mar 2022 · India

    Reaction-wheel attitude dynamics and HIL validation at ISRO’s Inertial Systems Unit (IISU).

    • Built and validated a three-axis reaction-wheel attitude dynamics model in MATLAB/Simulink; closed the sim-to-real loop on a hardware-in-the-loop bench with motor electronics.
    • Scripted sweeps to tune controllers and designed fixtures so vibe/thermal tests could be repeated exactly on flight-class reaction-wheel assemblies.
    • Stack: Simscape Multibody, quaternion kinematics, reaction-wheel torque models; EKF for rate estimation; PID controllers for step/slew.
    ISRO internship summary PDF thumbnail
    ISRO — Reaction-Wheel Attitude Dynamics (PDF)
    Modeling, HIL setup, controller tuning sweeps, and validation notes.
    Open PDF
    MATLAB/Simulink Simscape EKF HIL

  3. Research Fellow | Indian National Academy of Engineering (INAE)

    Apr 2021 – Aug 2021 · India

    Robotic welding optimization with feedback control and computer vision quality checks.

    • Tuned a robotic welding process using structured experiments; closed the loop with temperature feedback and automated weld quality checks with computer vision.
    • Design of Experiments (Taguchi) for factor screening; Response Surface Methodology for optimization. IR camera for bead temperature; PI control for adaptive wire-feed; PLC via Modbus.
    INAE research fellow PDF thumbnail
    INAE — Robotic Welding Optimization (PDF)
    DOE & RSM setup, feedback control loop, CV-based quality checks.
    Open PDF
    DOE/Taguchi RSM Computer Vision Control

  4. Research Intern | Indian Institute of Technology Kanpur (IIT Kanpur)

    Feb 2021 – Apr 2021 · Kanpur, India

    Structural/vibration analysis for a bouncing drop on a vibrating bath experimental setup; analytical models (SDOF→2-DOF→MDOF) and MATLAB/Python tooling.

    • Derived closed-form responses and frequency-response functions; performed modal decomposition and proportional damping analysis.
    • Implemented eigenanalysis, FRFs, and time-domain simulations to guide rig design and shaker excitation parameters.
    • Documented design constraints (instrumentation effects, FSI considerations, temperature drift) and validation plan.
    IIT Kanpur internship PDF thumbnail
    IIT Kanpur — Vibrating Bath / Droplet Dynamics (PDF)
    Analytical derivations (SDOF/2-DOF/MDOF), FRFs, MATLAB listings, and design notes.
    Open PDF
    Vibrations Modal Analysis MATLAB Python

    5. Project

    Collaborate →

    CEP Nexus v2 — Calibration & Policy Optimization

    CIS 614 • Harrisburg University

    A compact Climate–Economy–Population system-dynamics model upgraded with a two-reservoir carbon cycle, two-layer energy balance, and an abatement policy module (MAC curve). Implements parity between Insight Maker and Python for V&V, plus light calibration and policy search.

    System Dynamics Python Optimization Climate Modeling
    CEP Nexus v2 PDF thumbnail
    Final Project Progress Report (PDF)
    Model overview, equations, parameters, V&V plan, early results, and timeline.
    Open PDF

    Distributed Control System Architecture

    GitHub →

    Modular real-time control framework with lock-free IPC, zero-copy shared memory, and runtime plug-in modules for sensors, actuators, and control loops.

    C++17 POSIX IPC Zero-Copy Google Test CMake
    Distributed Control System — Architecture Overview
    Short demo: plug-and-play modules with sub-millisecond IPC and hot-swapping.

    Pythogoras Common Divisor Visualization

    Code → Demo →

    A lightweight HTML5-Canvas app that teaches the greatest common divisor visually. It compares two fixed segments (36 in and 10 in) using a movable probe segment (1 in or 2 in) and shows when both divide cleanly. Built for clarity on phones and desktops with crisp tick marks, animation, and guided steps.

    HTML5 Canvas Vanilla JS CSS3 Math EdTech
    Live Demo Open Fullscreen →

    Interactive Features

    • Toggle XY Length: switch probe between 1″ and 2″.
    • Show Divisions: tick marks and labeled multiples.
    • Animation Mode: auto-alternates probe length to compare cases.
    • Step-by-Step: guided checks for integer quotients on both lines.

    Why it works

    The visualization encodes the number-theory fact that any common divisor of 36 and 10 must divide their GCD, which is 2. Hence, only a 1″ or 2″ probe can land cleanly on both scales—making the GCD obvious at a glance.

    Technical Skills

    Robotics

    • ROS 2 / DDS, MoveIt 2, Gazebo, RViz
    • Perception (LiDAR, RGB‑D, IMU fusion)
    • Controls (PID, state estimation, Kalman)

    AI/Systems

    • Python, C++, CUDA, PyTorch/JAX
    • ONNX, TensorRT, Docker, CI/CD
    • Optimization, numerical methods

    Hardware/CAD

    • SolidWorks, CATIA V5, Fusion 360
    • ANSYS (FEA), KiCad/Altium
    • STM32, ARM Cortex‑M/A

    Publications

    1. Structural Topology Optimization of Headphones — PDF
    2. Design of Novel Fish De‑Scaler System — PDF
    3. Bi‑Directional Drone Design and its Path Planning — PDF

    Resume

    One‑page PDF tuned for robotics/AI roles.

    Download PDF

    Let’s build.

    Email soutrik.viratech@gmail.com or message on LinkedIn.

    Philadelphia, PA • Open to relocation