Asier Arranz | Senior Staff Developer Advocate @ NVIDIA

Asier Arranz

Senior Staff Developer Advocate @ NVIDIA

robotics & physical_ai

GitHub GitLab LinkedIn X Medium YouTube Email Website Resume

I make robots learn in GPU simulation and then prove it on real hardware, on whatever each demo needs: from an 8 GB Jetson to the biggest GPUs NVIDIA makes. I build the whole Physical AI pipeline: physics sim, models, language interface, and the ops to keep it alive on a show floor. My format is the live demo: real hardware, running in front of the audience.

I joined NVIDIA in 2020 and since October 2024 I am Senior Staff Developer Advocate in a global role. Before NVIDIA I led IBM's Global Quantum Community Lab (I hold a patent for 3D quantum circuit visualization), was an Oculus Launchpad Fellow at Meta, and ran the Technology Lab at IE Business School. In 2019 I built AskSkynet.com, the first public web interface to chat with a GPT model, years before ChatGPT existed.

what_i_do_at_nvidia

Physical AI demos: humanoids and robot arms simulated with Newton and Isaac Sim/Lab, connected to language interfaces, and operated live at international NVIDIA events.
Robot learning: GR00T foundation models, LeRobot SO-100/101 arms, imitation learning, and XR teleoperation with CloudXR.
Generative AI on edge: the latest models (Qwen, Gemma, Llama, multimodal VLMs) running on Jetson Orin and Thor, squeezing real performance out of every watt.
Teaching: workshops and hands-on labs that take a developer from zero to a working demo on their own hardware.

highlighted_work_2026

Physical AI, live in Japan A Unitree G1 humanoid walking through a simulated smart factory and a Franka arm stacking cubes, physics fully on GPU with NVIDIA Newton, responding to commands in English, Spanish and Japanese. Built end to end and handed over to the NVIDIA team in Japan, where it runs today.

Teleoperating robots in VR For the GR00T N1.7 + LeRobot launch with Hugging Face, I drove the SO-101 arm inside Isaac Lab from a Quest 3 headset, and fixed the streaming stack along the way with two upstream contributions to Isaac Teleop (Seeed article).

Two humanoid robots, one Jetson NemoCast: a live AI news show hosted by two UBTECH Alpha Mini robots at Computex 2026, with Gemma 4, Kokoro TTS and live English to Mandarin translation running locally on a single 8 GB Jetson Orin Nano Super.

Watching a world model think Interactive NVIDIA Cosmos demos on 8x H100s: real footage next to generated video on a slider, and a live viewer that shows every denoising step while the model generates.

Text to video on a desktop box A complete AI video pipeline (LTX-2.3, 22B parameters) built for GTC Taiwan, running entirely on one DGX Spark: type a prompt, get a finished clip.

code

github.com/asierarranz: personal projects and demos, from Gemma on Jetson to GPU benchmarks.
gitlab.com/asier_nvidia: public releases of my NVIDIA demos, including the Physical AI demo running in Japan.
Upstream: two pull requests to NVIDIA Isaac Teleop stabilizing Quest 3 streaming for VR teleoperation (PR #726 and #753), sent for the GR00T N1.7 + LeRobot launch.