NewsNovember 17, 2026Редакция 105.kz
Pick-and-place robot simulation on SimReady assets: Hugging Face breakdown
Hugging Face showed a step-by-step methodology for simulating a can pick-and-place with a UR17G robot and Robotiq gripper in Isaac Sim. We examine what this offers to industrial enterprises in Kazakhstan.
Key takeaway. The Hugging Face blog described a pick-and-place simulation example: a UR17G robot with a Robotiq 2F-140 gripper moves a can in Isaac Sim using SimReady assets. For a plant, this is a way to validate geometry, contacts, and actuator limits before going to the real line. The methodology is applicable as a stage of a digital twin of a robotic cell, but the results are currently limited to a fixed scene and do not replace tests on equipment.
What happened
In the Hugging Face blog, a breakdown of a pick-and-place robotic cell simulation was published. The example uses a UR17G robot arm, a Robotiq 2F-140 gripper, and a tin can model. The scene is assembled in Isaac Sim 6.2, and agent-generated code is used for motion and grasping control; the author mentions GPT-6 Sol as one option, but notes that it can also be done manually.
Assets are loaded from the SimReady Central library. SimReady means the OpenUSD model is prepared for a specific simulation scenario: it has collisions, mass, joints, and actuators defined. Validation is performed against profiles: for the arm — Robot-Body 4.0.0, for the gripper — Robot-Gripper 3.0.0. The reports record successful checks of PhysX actuators. After assembly, behavior is verified: joints, contact of the can with the surface, holding in the gripper during movement.
The source emphasizes: the results apply only to a fixed scene. The can position is known, vision is not used, can deformation and liquid motion are not modeled, gripping force is limited to 2 N·m and not calibrated against the gripper datasheet. This is a methodological demonstration, not proof of robustness to position variations or to real equipment.
What this means for enterprises in Kazakhstan
For food, packaging, and machine-building plants in Kazakhstan, such examples are useful as a template for validating robotic operations before procurement or installation. Typical operations include placing cans, bottles, boxes, loading machines, palletizing. Simulation allows checking whether the arm reaches specified points, does not hit safety barriers, and whether the gripper holds the object during acceleration and deceleration.
For enterprises with legacy process control systems, the ability to import CAD models and kinematics of real robots is important. The example uses ready SimReady assets, but for local equipment, models often need to be built from drawings or 3D scans. This is labor-intensive, but the result provides a basis for a digital twin of the cell — the robot-gripper-object-work surface linkage. Such a twin can be used for operator training and for testing layout changes without stopping the line.
Separately, the on-premise mode should be noted. Simulation in Isaac Sim can run on a local workstation or enterprise server; CAD data and process parameters do not leave the perimeter. If an external AI agent is used for code generation, it is necessary to evaluate where data is sent. In sensitive production environments, the scenario can be assembled manually or using local AI (assessment). Language issue: documentation and interface are in English; engineering teams in Kazakhstan generally cope with this, but for knowledge transfer in Russian and Kazakh, methodological materials will need adaptation.
How to apply: 8–12 week pilot
A pilot project based on this methodology could look like this:
- Weeks 1–2. Audit of one robotic operation: layout, CAD models, robot and gripper datasheet data, actuator limitations, working zones, occupational safety requirements. Define a target metric — for example, cycle time, placement accuracy, absence of collisions.
- Weeks 3–4. Preparation of models. If there are no ready SimReady assets for the equipment, create simplified OpenUSD models with correct collisions, masses, and joints. Validate them in the scene.
- Weeks 5–6. Scene assembly and joint tuning. Check angle limits, actuator settings, contact of the object with the surface. Ensure the model does not fall through or fly apart.
- Weeks 7–8. Run the pick-and-place scenario with fixed points. Measure grasp, transfer, and placement success; record positioning errors. Compare with calculated values.
- Weeks 9–12. Evaluate results and document limitations. If the effect is positive, plan the next step: adding machine vision, a test on a real robot with trajectory transfer, or personnel training.
This loop aligns with the approach described on the Digital Twins page and is applicable at Industrial Facilities. It is important not to proceed to physical startup until collisions are eliminated and object holding is confirmed in simulation.
Limitations and risks
The main limitation of the source is the narrowness of the scenario. The authors explicitly state: fixed starting position, no machine vision, a simplified can model without deformation and liquid sloshing, scripted motion along fixed points without replanning. The gripping force of 2 N·m is chosen in simulation, not from manufacturer documentation. Therefore, transferring results to a real workshop requires separate validation.
For Kazakhstani plants, additional risks: inaccurate CAD data of legacy equipment, differences between the model and actual kinematics after repairs, lack of ready SimReady assets for local robots. Manual model refinement and calibration against real measurements will be needed. If an external AI agent is used for code generation, it is necessary to check whether drawings and process parameters are leaving the enterprise. For industrial operation, it will likely be necessary to move the simulation into a closed loop and integrate with the production management system (assessment).
Thus, the Hugging Face material is useful as an engineering recipe, but not as a ready solution for a specific enterprise. Its value lies in demonstrating the order of validating joints, contacts, and control code before physical startup.
Sources
- Build a Robot Pick-and-Place Simulation with SimReady Assets · 2026-11-13T18:00:39.000Z
SimReady means an OpenUSD asset has been prepared with the properties needed for a defined simulation use case, such as collisions, mass, joints, and joint drives.
- Иллюстрация: сгенерирована 105.kz (z-image, Apache 2.0)
This material was prepared with AI from the sources listed above. Spotted an inaccuracy? Write to info@105.kz.