Hands-On Humanoid Robot Simulation With NVIDIA Isaac GR00T and Newton
This tutorial is part of the RIVF 2026 programme. It runs as a half-day, simulation-only laboratory of four hours for up to 60 learners, delivered by NVIDIA. See Tutorials for the full list of accepted tutorials.
| Tutorial Date | December 18, 2026 |
|---|---|
| Tutorial Time | To be confirmed |
| Conference Dates | December 18–20, 2026 |
Abstract
A four-hour, simulation-only laboratory for researchers, graduate students and robotics engineers. The first two hours adapt NVIDIA's AI Day Singapore Isaac GR00T lab: learners inspect a Unitree G1 apple-to-plate scene in Isaac Sim and Isaac Lab-Arena, work with demonstration data, and evaluate a GR00T 1.7 policy. The next two hours use Newton Fundamentals to build and control physics simulations. The GR00T G1 scene uses PhysX; Newton is taught as a separate hands-on track.
Audience and Prerequisites
Graduate students, faculty, R&D and robotics developers. Participants need intermediate Python, Linux terminal and Docker experience, and basic machine learning concepts; robotics or simulation experience helps. No physical robot is required.
Learning Objectives
- Validate a G1 simulation.
- Trace demonstrations through LeRobot to GR00T policy evaluation.
- Run Newton simulation, robot control, inverse kinematics and coupled manipulation exercises.
Schedule
The session is split evenly between the two modules. Times below follow the proposed 08:30–12:30 start; the exact start and end time are to be confirmed with the organizers.
Module 1 — Isaac Sim and GR00T simulation (120 minutes)
| Time | Activity and participant output |
|---|---|
| 08:30–08:45 | Introduce the simulation-only G1 pick-and-place workflow and the scene's success criterion. |
| 08:45–09:10 | Launch the prepared Isaac Sim and Isaac Lab-Arena environment; validate the G1 scene and inspect task configuration. |
| 09:10–09:35 | Observe instructor teleoperation and whole-body control; inspect recorded HDF5 demonstrations and observations. |
| 09:35–09:55 | Convert a provided demonstration sample to LeRobot format; review GR00T 1.7 post-training configuration. |
| 09:55–10:20 | Run a supplied GR00T checkpoint in closed-loop Arena evaluation; compute success rate and inspect failures. |
| 10:20–10:30 | Review outputs, questions and transition to Newton's separate environment. |
Provided data and a trained checkpoint make live evaluation achievable in two hours. Full GR00T post-training is explained but is not a timed participant exercise.
Module 2 — Newton Fundamentals (120 minutes)
| Time | Activity and participant output |
|---|---|
| 10:30–10:40 | Open the prebuilt Newton lab; identify ModelBuilder, State, Control, Contacts and Solver. |
| 10:40–11:05 | Build a box-and-sphere scene, step physics and compare execution with CUDA graph capture. |
| 11:05–11:30 | Load a Franka arm; apply joint forces and position targets, then compare motion. |
| 11:30–11:55 | Use inverse kinematics to drive an end-effector target and inspect the resulting pose path. |
| 11:55–12:20 | Run a coupled rigid-arm and deformable-cable manipulation example and inspect solver behaviour. |
| 12:20–12:30 | Review runnable scripts and results; questions and next steps. |
Take-home outputs: runnable Newton lesson scripts, a G1 rollout and success-rate summary, slides, a reading list and troubleshooting notes.
Hands-on Component
The lab is planned for up to 60 attendees working in 30 pairs, with a lead instructor and two facilitators. NVIDIA will prepare and rehearse separate GR00T and Newton environments, pre-stage the dataset, checkpoint and lesson assets, and share participant instructions. The GR00T simulation path calls for Linux, Docker and an RTX-class GPU; live post-training needs at least 48 GB of VRAM, so it is demonstrated through configuration rather than completed in the session.
Instructors
Herman Wu — Principal Instructor, NVIDIA Deep Learning Institute
Technical presenter and training supporter.
Dr. Quan Tran Minh — Senior Developer Relations Manager, Asia Pacific South, NVIDIA
Corresponding author and local programme lead.
Module 1 is based on NVIDIA's four-hour GR00T hands-on lab at AI Day Singapore (22 September 2026); Module 2 follows NVIDIA's two-hour Newton Fundamentals curriculum. The final instructor and teaching assistant roster is to be confirmed.
Contact
For inquiries, please contact: rivf2026@vinuni.edu.vn