NVIDIA Omniverse & Isaac Sim Developer — Digital Twin & Simulation
I build custom Omniverse Kit extensions and industrial digital-twin simulations on OpenUSD — physics, robotics, and 3D reconstruction.
Scroll to fly through the facility
Cell 01 Gaussian splat
Images and video → simulation-ready 3D you can rotate right here.
Cell model: “Modern Coastal Hillside Villa” by Visthétique (Sketchfab), CC-BY 4.0. Modified: re-rendered and reconstructed as a 3D Gaussian Splat.
View project detailsCell 02 Pick & place
Manipulation on a factory twin.
Cell 03 Contact heatmap
See exactly where a structure takes the load.
Cell 04 FEM rope
Stable rope physics for crane lifting.
Cell 05 Tote line
Totes routed through a live conveyor line.
Cell 06 Motion Path Kit
Drive it, record it, replay it.
Cell 07 Synthetic data
One simulation, infinite training scenes.
The facility
Every cell above was simulated live in Isaac Sim, baked to USD time samples, composed into one stage and rendered with physics off. The model you just flew through is exported from that same stage.
02
I'm a simulation and real-time 3D developer with 3+ years across XR and digital twins, and around 1.5 years hands-on with NVIDIA Omniverse / Isaac Sim. I design and ship custom Omniverse Kit extensions on a clean, layered architecture — from an engine-agnostic, unit-testable backend through isolated UI — covering real-time physics analysis, motion-path authoring, and 3D reconstruction.
I've delivered industrial digital-twin work for manufacturing and logistics: robot-arm pick-and-place, crane and rope heavy-lift with deformable physics, and condition-monitoring data pipelines. Before Omniverse, I built industrial AR on HoloLens 2 with on-device computer vision. Alongside my full-time role, I run a Top Rated freelance practice on Upwork (100% Job Success, $20K+ earned) in XR and simulation.
03
04 · Featured Work
Images and video → simulation-ready 3D you can rotate right here.
An Omniverse Kit extension that turns photos or video into 3D Gaussian splats and refined meshes — with text-prompt object isolation (SAM3), point-cloud-to-mesh tuning, and USD/USDZ export ready for simulation. Docker-orchestrated GPU backends (FVDB, COLMAP, Kaolin) with hardware validation and resumable pipelines. Iterated across six monthly releases.
[ p1_splat_viewer.html ]
Runs in your browser · needs WebGL
Totes routed through a live conveyor line.
The rollers of NVIDIA's own warehouse conveyor asset are the drive surface: an IsaacConveyor OmniGraph node per roller body, with velocities baked in world frame so the line keeps moving even with the extension disabled. Each tote is a rigid body whose collision proxy is separate from its visual mesh, drawn from a pre-built pool so no physics prim is created or destroyed while the timeline runs. A single divert point pushes a tote off the line into one bin. Captured headless offline on NVIDIA's public sample warehouse assets — 1800 frames, audited at zero black frames.
Drive it, record it, replay it.
Author operational motion paths for vehicles, AGVs and forklifts by driving them in a third-person view, then record sparse, editable 3D waypoint tracks and replay them with a progress-parameterized follow controller — a blocked body waits rather than skips. Includes a Pure-Pursuit steering model for articulated forklifts with rear-wheel-steer kinematics.
See exactly where a structure takes the load.
A dedicated extension for real-time contact analysis on mesh and 3D-Gaussian-Splat assets. Two per-vertex heatmap engines — Euclidean-distance and a topological mesh-adjacency (CSR-graph heat-diffusion) engine — run on NVIDIA Warp GPU kernels, with scale-aware re-baking and floating force-labels marking each hot zone.
Manipulation on a factory twin.
Factory robot-arm pick-and-place built on Isaac Sim's manipulator (RMPflow) example framework: a rigged arm with a suction end-effector runs a multi-phase state machine to move a part from A to B along a collision-aware RMPflow trajectory, with an OmniGraph-driven conveyor. Shown here as a clean demo on NVIDIA's stock factory and robot-arm assets.
Stable rope physics for crane lifting.
Deformable-rope physics for heavy-lift simulation: after root-causing why rigid cylinder-and-joint rope chains go unstable under load, I moved to a unified deformable (FEM) rope on PhysX 5 and built a Kit extension to tune stiffness, damping and density in real time. Part of a networked crane heavy-lift platform delivered for an industrial client.
One simulation, infinite training scenes.
Domain-randomization showcase: a robot pick-and-place task runs continuously while the scene re-randomizes several times a second — environments, lighting, object shapes, colors, scales and camera viewpoints — the way synthetic training data is generated for robot-learning and vision pipelines. Captured fully headless in Isaac Sim with a scripted cinematic-capture harness on NVIDIA's manipulator example framework.
05
06
Shipped 3 Omniverse Kit extensions on a clean layered architecture; delivered industrial digital-twin simulations (crane/rope, pick-and-place, condition monitoring). Began with a HoloLens 2 AR + on-device-AI app, then moved to Omniverse/Isaac Sim.
100% Job Success, $20K+ earned across 6 client contracts.
VR/AR/MR industrial apps on Meta Quest 2/3 and HoloLens 2.
07
Open to Omniverse / Isaac Sim, digital-twin, and simulation roles — full-time or freelance.
