NVIDIA walkthrough shows robot pick-and-place simulation with SimReady assets

What the walkthrough covers

NVIDIA has published a tutorial on the Hugging Face Blog that shows how to build a robot pick-and-place simulation in Isaac Sim, the company's robotics simulation platform. The example task uses a UR17G robot arm, a Robotiq 2F-140 gripper and a beverage can.

The guide, published on 13 November 2026, walks through downloading the assets, assembling the robot, validating its physics and running the task with controls generated by an AI agent.

What SimReady means for an asset

In this context, SimReady describes an OpenUSD asset that has been prepared with the properties a specific simulation needs. Those properties can include collision geometry, mass values, joints and joint drives, which are the settings that make a motorised joint behave correctly.

For the pick-and-place task, this covers the arm and gripper joints plus the can's collisions and mass.

The assets come from SimReady Central, an open library hosted on Hugging Face. It offers robot and object models with authored simulation properties and evidence that they passed profile validation. A search service endpoint also lets users look for content curated by original equipment manufacturers.

How the scene is built

The asset packages include the models and their simulation settings, but the user supplies the rest. The scene needs task code that defines the arm and gripper as one assembly, a work surface, starting positions and the commands that move the arm and open and close the gripper.

The tutorial uses Isaac Sim 6.2 and an AI agent to help create and run the task code. The example pairs the platform with GPT-6 Sol, though NVIDIA notes that other agent and model combinations may work if they can load the required skills and run Python against Isaac Sim. Results from generated code can vary, so the company advises inspecting the code and checking the behaviour before adapting the task.

To let the agent interact with the simulation, Isaac Sim is started with its remote Python server enabled. The agent then uses bundled skills such as isaac-sim-remote and usd-articulation to assemble the arm and gripper and validate the physics, treating the pair as a single controllable unit.

Testing before the real task

NVIDIA warns that a robot can look correct while stationary yet move or interact badly once the simulation runs. Before attempting the pick-and-place task, the guide recommends three checks: reviewing the joint limits and drive settings, testing a small arm movement and the gripper opening and closing, and confirming the can stays on the surface rather than passing through it.

During the run, the key thing to watch is whether the can remains held in the gripper as the arm moves. Because asset structures differ, NVIDIA suggests using the settings in the loaded scene as a starting point rather than copying values blindly.

What validation reports show

The example assets ship with validation reports. The arm uses the Robot-Body 4.0.0 profile and the gripper uses Robot-Gripper 3.0.0. Both reports record passing PhysX driven-joint checks, which inspect the joint-drive settings.

According to the blog post, this profile validation gives a shared baseline for reusing the assets across projects. The pick-and-place run then shows how the arm, gripper, can and controller behave together once assembled, which is the step that matters most in practice.

Why it matters

Preparing robot models for physics simulation has traditionally required manual work on collisions, mass and joint behaviour. SimReady Central aims to remove that step by offering assets that arrive with these properties already authored and verified against published profiles.

For teams working on robot learning or automation, the walkthrough also shows a shift in workflow: an AI agent, rather than a programmer, can assemble the scene and write the control code, with the human checking the results. NVIDIA cautions that generated code and outcomes can vary, so verification remains part of the process.