← News

ResearchSources: huggingface.co · 13 Nov 2026

NVIDIA walkthrough shows agent-built robot pick-and-place in Isaac Sim using SimReady assets

A new tutorial combines a UR17G arm, Robotiq 2F-140 gripper, and beverage can from SimReady Central, with an AI agent generating the task code.

In plain words

NVIDIA published a tutorial for building a simulated robot that picks up a beverage can and places it elsewhere. The robot parts and can come from SimReady Central, a library of 3D assets pre-configured with physics settings, and an AI agent writes the code that runs the task in the Isaac Sim simulator.

What was announced

The walkthrough, published on Hugging Face, uses the UR17G arm and Robotiq 2F-140 gripper alongside a beverage can, all sourced from SimReady Central, an open asset library NVIDIA describes as containing OpenUSD assets prepared with simulation properties such as collisions, mass, joints, and joint drives. The authors used GPT-6 Sol as the agent to generate and run task code, noting that other agent-model combinations may work and that generated results can vary. Before running the task, the guide advises checking joint settings, testing small movements, and confirming the can rests on the surface without falling through.

The validation reports for the arm and gripper used the Robot-Body 4.0.0 and Robot-Gripper 3.0.0 profiles respectively, both recording passing PhysX driven-joint checks. NVIDIA is explicit that the demonstration is narrow: the can starts at a known position, no vision is used, collision checks are limited, and the can is simplified with no deformation or liquid motion. The 2 N·m drive-effort cap for the grasp was chosen through simulation tests, not calibrated to fingertip forces or manufacturer recommendations. The authors state the results apply only to this scripted setup and do not establish robustness to uncertain object positions, different payloads, or real hardware.

What it does not show

  • The article is a single-source walkthrough by NVIDIA; no independent party has reproduced the pick-and-place results.
  • The example uses a fixed scene with known object positions, no vision, limited collision checks, and a simplified can, so it does not test real-world robustness.
  • The agent-generated code and results can vary, per the authors, so outcomes are not guaranteed to match the walkthrough.

For security and IT teams

Scope
The workflow requires Isaac Sim 6.2, an agent that can use the bundled Isaac Sim skills, and assets downloaded from SimReady Central on Hugging Face.
Status
The tutorial is published and the assets are available for download from SimReady Central, with validation reports for the arm and gripper included.

Sources

Source-support review completed 13 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.

Updates and corrections

No corrections so far.

First published 13 Nov 2026

New to the subject? AI from the Ground Up starts from the beginning.