Pranav Kolekar - Robotics Software Engineer

Pranav KolekarRobotics Software Engineer | Physical AI & Sim-to-Real

Get In TouchContact

About Me

Building Physical AI systems through cloud simulation, digital twins, and sim-to-real robotics deployment.

Professional Summary

Current Role

Robotics Software Engineer at Neural Foundry, building cloud-scale simulation and sim-to-real robotics systems.

Education

Bachelor's in Electronics & Telecommunications, M.E.S Wadia College of Engineering, Pune

Specialization

Physical AI, NVIDIA Isaac Sim, ROS 2 systems, simulation infrastructure, and cloud automation.

NVIDIA Isaac Sim digital twins

ROS 2 Nav2 and DDS QoS tuning

Robot workflow integration

Cloud GPU simulation infrastructure

Robotics Software Engineer driving the evolution of Physical AI through cloud-scale simulation, automated digital twins, and robust sim-to-real deployment pipelines.

Specialized in connecting robotics middleware, device interfaces, and photorealistic simulation environments (NVIDIA Isaac Sim) to accelerate validation from cloud GPU systems to physical hardware.

Download Resume Get In Touch

Pranav Kolekar - Robotics Software Engineer

Cloud Simulation

Automated reproducible simulation deployments with Brev Launchables and Docker Compose

Digital Twins

Built Isaac Sim environments for autonomous mobile manipulation and validation

Sim-to-Real Control

Built workflows that keep simulation behavior aligned with production robot execution

Simulation Reliability

Debugged simulator-side failures and shipped repeatable team-level workarounds

Experience & Education

Robotics software experience across Physical AI simulation, embedded control, ROS 2 systems, and autonomous deployment.

Work Experience

Robotics Software Engineer

Full-time

Neural Foundry

Hyderabad, India

Feb 2026 – Present

Isaac Sim, ROS 2, mobile manipulation, and embedded C++ — building a cloud-based digital twin for lab-automation hardware and operationalizing an AI-assisted development workflow across the simulation stack.

Selected Work

Architected a custom NVIDIA Isaac Sim C++/Python extension bridging embedded firmware with an 8-device lab-automation digital twin, managing MQTT connection lifecycle across all four simulation states (play/pause/step/stop) for firmware-in-the-loop testing entirely in the cloud

Engineered a Docker Compose cloud simulation environment (AWS + L40S GPU) that collapsed a 6-step manual provisioning process into a single reproducible deployment; resolved 2 blocking GPU instance lifecycle failures

Built an end-to-end mobile manipulation digital twin (differential-drive AMR + 6-DOF cobot) in Isaac Sim with full ROS 2 / ros2_control integration; tuned Nav2 across 5 subsystems to enable autonomous vision-guided pick-transport-place

Diagnosed a confirmed NVIDIA Isaac Sim 5.1.0 bug — a vacuum gripper surface-joint baking a world-frame anchor into USD at authoring time, causing the articulation to fly off on reference spawn — via OpenUSD Pxr layer inspection; authored the workaround, NVIDIA-acknowledged, fix targeted for 6.x

Benchmarked multi-instance Isaac Sim scalability on NVIDIA L40S GPUs, quantifying a 15× FPS drop (120 → 8 fps) across 8 concurrent instances and isolating a ~4-core-per-instance CPU ceiling

Converted continuous AprilTag/ArUco detection to an on-demand ROS 2 service, restoring production camera streaming from under 6 Hz to a consistent 15 Hz via targeted DDS QoS tuning

Technologies Used

NVIDIA Isaac SimROS 2OpenUSDDockerMQTTNav2ros2_controlEmbedded C++

Robotics Software Engineer

Full-time

Nxtwave Disruptive Technologies

Hyderabad, India

Jun 2025 – Jan 2026

ROS 2, Autoware, manipulation, and imitation learning — extending an SDK for the Unitree Go2, building a manipulator control stack, and deploying Autoware on an F1TENTH vehicle.

Selected Work

Extended an existing ROS 2 SDK for the Unitree Go2 quadruped, authoring launch files and enabling dual-transport connectivity over WebRTC and CycloneDDS

Implemented a serial communication and control stack for a 6-DOF manipulator driven by bus servos; derived DH parameters and computed forward kinematics for real-time end-effector pose estimation

Established a leader–follower teleoperation pipeline using LeRobot; collected 50+ demonstration episodes and trained ACT and Diffusion Policy models on CUDA GPUs, achieving repeatable task execution on physical hardware

Deployed the full Autoware autonomous driving stack on an F1TENTH vehicle and validated end-to-end indoor autonomous navigation including NDT localization, lane-keeping, and pure-pursuit path tracking

Produced Autoware-compatible vector maps using Tier IV Map Builder and processed raw LiDAR point clouds in CloudCompare through statistical filtering and dataset merging

Technologies Used

AutowareROS 2LeRobotACTDiffusion PolicyF1TENTHCUDA

Robotics Developer

Contract

Ondroid

Finland (Remote)

Feb 2025 – Present

Automated computer vision pipelines and CI test coverage for a robotics assessment platform.

Selected Work

Developed Python/OpenCV automated computer vision pipelines for output verification, eliminating manual review and standardizing evaluation outcomes

Integrated Pytest into the CI backend pipeline, increasing test coverage and enforcing regression-free deployments across multiple release cycles

Technologies Used

PythonOpenCVPytestCI/CD

Robotics Lead — Robocon

Leadership

M.E.S. Wadia College of Engineering

Pune, India

Feb 2024 – Jul 2024

Led hardware and software R&D for a two-robot competition team, representing the college at ABU Robocon 2024.

Selected Work

Led hardware and software R&D for a two-robot team, integrating closed-loop motor control and custom control firmware on Robot 1

Provisioned a complete ROS 2 navigation stack on Robot 2 — URDF modeling, wheel encoder odometry, 2D LiDAR, IMU integration, and Gazebo simulation for pre-competition testing

Represented the team at ABU Robocon 2024, New Delhi

Technologies Used

ROS 2GazeboURDFLiDARIMU

Electronics Intern

Internship

Canspirit AI

Pune, India

Mar 2024 – May 2024

IoT sensor integration and cross-platform data logging for real-time acquisition pipelines.

Selected Work

Connected IoT sensor nodes to Arduino Cloud for real-time data acquisition

Built cross-platform Python logging pipelines to streamline data storage, retrieval, and downstream analysis

Technologies Used

IoTArduino CloudPythonEmbedded Systems

Embedded Systems Developer

Project-based

Indkarta LLP

Pune, India

Oct 2023 – Jan 2024

PCB design and firmware development for embedded systems in clinical medication-processing environments.

Selected Work

Designed custom PCB layouts in KiCad for two embedded systems deployed in clinical medication-processing environments

Owned firmware development from prototype through field testing

Technologies Used

FirmwarePCB DesignKiCad

Education

Bachelor of Engineering

Electronics and Telecommunications

M.E.S. Wadia College of Engineering

Pune, Maharashtra · Dec 2021 – May 2025 · First Class

Electronics and telecommunications with a focus on embedded systems and robotics.

Digital Signal ProcessingRobotics & AutomationEmbedded SystemsControl Systems

Systems I've Shipped

Real debugging and infrastructure work from production robotics systems — root causes, not just outcomes.

The articulation that flew off on spawn

A mobile manipulator was stable in its own Isaac Sim world, but exploded on Play whenever it was spawned as a reference into another world.

→ Root cause isolated via direct Pxr/USD layer inspection, not trial-and-error→ NVIDIA-acknowledged defect, fix targeted 6.x

NVIDIA Isaac SimOpenUSDPhysXPython

Read the case study →

A digital twin for pick-transport-place

An end-to-end mobile manipulation twin — differential-drive AMR plus a 6-DOF cobot — built in Isaac Sim with full ROS 2 control integration.

→ 5 Nav2 subsystems tuned for reliable navigation with an arm mounted→ Full ROS 2 / ros2_control integration across AMR + 6-DOF cobot

NVIDIA Isaac SimROS 2Nav2ros2_controlAprilTag/ArUco

Read the case study →

Six manual steps into one deployment

Turned a slow, error-prone manual Isaac Sim provisioning process on cloud GPUs into a single reproducible Docker Compose deployment.

→ 6-step manual process collapsed into 1 reproducible deployment→ 2 blocking production infrastructure issues resolved

Docker ComposeAWSNVIDIA L40S GPUNVIDIA Isaac Sim

Read the case study →

Trading continuous detection for a service call

Converted a continuously polling AprilTag detection node into an on-demand ROS 2 service, then chased down an unrelated camera-streaming regression it exposed.

→ Continuous detection converted to on-demand ROS 2 service→ Streaming recovered from <6 Hz to a consistent 15 Hz via DDS QoS tuning

ROS 2AprilTag/ArUcoWebRTCDDS QoS

Read the case study →

How many simulations fit on one GPU

Benchmarked how Isaac Sim performance degrades as concurrent simulation instances scale on a single cloud GPU, to inform infrastructure sizing.

→ 120 → 8 fps across 8 concurrent instances on an L40S GPU→ ~4-core-per-instance CPU ceiling isolated as the binding constraint

NVIDIA Isaac SimNVIDIA L40S GPUDocker

Read the case study →

Technical Skills

Production robotics stack, plus the perception/ML track actively in progress — cyan-dimmed skills are being built, not yet shipped.

Programming Languages

Core languages across robotics, embedded, and tooling work.

PythonC/C++MicroPythonBash

Robotics Frameworks

Navigation, manipulation, and autonomy stacks used in production.

ROS 2 (Humble/Jazzy)Nav2ros2_controlMoveIt 2AutowareLeRobot

Simulation & Visualization

Digital-twin and debugging tooling for sim-to-real work.

NVIDIA Isaac SimGazeboOpenUSDRViz 2Foxglove Studio

Perception, SLAM & Mapping

Classical CV and multi-sensor mapping — the base the ML track builds on.

RTAB-MapCartographerGMappingOpenCVPCLAprilTag / ArUcoSensor Fusion (LiDAR, IMU, depth)

Robot Learning & Perception ML

Imitation learning shipped on physical hardware; 3D/deep perception ML is the active build.

Imitation Learning (ACT, Diffusion Policy)CUDAPyTorch· building3D Object Detection (PointPillars, CenterPoint)· buildingSensor Fusion BEV (BEVFusion)· buildingTensorRT / TAO Deployment· buildingOpenPCDet· buildingMonocular Depth Estimation· buildingDomain Randomization (Sim-to-Real Transfer)· buildingKITTI / nuScenes Benchmarks· building

Infrastructure & Tools

Cloud, comms, and delivery tooling around the robotics stack.

DockerAWSMQTTFastAPIGitPytestKiCadURDF/SDF

Hardware & Interfaces

Physical platforms and interfaces worked with directly.

ODrive S1CAN busUnitree Go2Doosan CobotsNVIDIA JetsonRaspberry PiBus Servos

Open Source Contributions

Contributing to the ROS 2 Navigation2 ecosystem and maintaining personal robotics tooling.

Nav2 Minimal TurtleBot Simulation — Bug Fix

ros-navigation/navigation2

Merged

Diagnosed and fixed a LiDAR sensor misconfiguration where rays at maximum range generated false floor collision detections (Issue #26), restoring correct obstacle detection for autonomous navigation testing.

C++

nav2_pose_saver

pranavk-2003/nav2_pose_saver

Owner

Lightweight ROS 2 C++ node for Nav2-integrated pose persistence in industrial environments — atomic file-swapping to prevent data corruption, periodic pose saving, and service-based dynamic control. Fully tested on ROS 2 Jazzy.

C++