Isaac Web Exporter

A robotics simulation contains more information than a fixed video can show. Reviewing a pick-and-place attempt means understanding the scene, the robot’s motion, the object being handled and the result of the attempt.

At the 2026 HPE & NVIDIA Agentic AI Hackathon in Geneva, I worked on a simulated cheese-factory workflow. Camera input is used to classify each cheese item and determine its destination; a Franka Panda robotic arm then attempts the corresponding pick-and-place operation.

My Physical AI work centered on the Franka arm in NVIDIA Isaac Sim. I worked with its articulated hierarchy and link motion, the moving items and the outcome of each attempt, so the workflow could be captured and reviewed beyond a final success frame.

Isaac Web Exporter brings that recorded simulation into a self-hosted Three.js experience. The export includes factory geometry, materials, robot and object motion, and guided run outcomes.

Simulation: NVIDIA Isaac Sim provides the 3D environment for the Franka Panda and the camera-based classification and manipulation workflow.

Export: Scene geometry and materials are packaged with recorded motion and run outcomes for browser playback.

Review: The Three.js viewer lets reviewers pause, scrub, step through frames, orbit the camera and inspect objects. They can follow classification and manipulation attempts without running Isaac Sim, Docker or an RTX server.

This is playback of a recorded simulation, rather than live control of a physical robot. The work described here uses a simulated Franka Panda.

I placed second at the 2026 HPE & NVIDIA Agentic AI Hackathon. The project combines robotics simulation with web development to make a Physical AI workflow easier to inspect and share.

Explore more of my work in software development, applied AI and interactive experiences.