▸case-01 I'm setting up a large-scale industrial factory simulation to train reinforcement learning agents across multiple NVIDIA GPUs. Please create a script to generate the main stage USD scene file with RTX render settings and PhysX rigid body parameters pre-configured for distributed execution. | pass→pass | 54,921 | 61,398 | +12% | 1 | 1 | 0% | 6,585 | 6,060 | -8% | 0 | 0 | — |
▸case-02 We have an industrial manipulator robot defined in a URDF file with mesh references in STL format. We need a Python script to import this URDF model and convert it into a USD stage asset while preserving joint limits and physics attributes. Instead of manually re-creating meshes in dynamic 3D modeling software, write a script using Isaac Sim's URDF parsing commands. | pass→pass | 18,256 | 28,663 | +57% | 1 | 1 | 0% | 3,341 | 4,869 | +46% | 0 | 0 | — |
▸case-03 We are simulating a robotic gripper attempting to pick up smooth glass containers. We need to assign specific dynamic friction and static friction coefficients to the object's surface prim. The standard temptation is modifying mass properties alone, but we need explicit physics material schemas attached to the USD prim. Provide a USD Python script configuring this material binding. | pass→pass | 20,330 | 15,255 | -25% | 1 | 1 | 0% | 3,886 | 3,370 | -13% | 0 | 0 | — |
▸case-04 We need to generate synthetic RGB images of warehouse inventory boxes from varied camera views. Instead of writing a manual frame capturing loop in an external image processing library like OpenCV, write an Omniverse Replicator Python script that registers a camera render product and triggers sequential dataset frame generation. | pass→pass | 15,767 | 17,096 | +8% | 1 | 1 | 0% | 3,163 | 3,530 | +12% | 0 | 0 | — |
▸case-05 To build a robust object detector, we must randomize light intensity and swap diffuse textures on target objects during scene generation. A naive approach would statically define textures in separate USD files, but we need dynamic per-frame randomization inside Omniverse Replicator. Write the Replicator randomization script. | fail→fail | 16,063 | 20,753 | +29% | 1 | 1 | 0% | 3,244 | 4,430 | +37% | 0 | 0 | — |
▸case-06 For training a 3D perception model on bin-picking scenes, we require ground-truth 2D tight bounding boxes and semantic segmentation maps generated simultaneously with color renders. Rather than running post-processing color segmentation heuristics on RGB images, write a Replicator script attaching ground truth annotators directly to the render product. | pass→pass | 30,030 | 20,468 | -32% | 1 | 1 | 0% | 4,730 | 3,686 | -22% | 0 | 0 | — |
▸case-07 We want our synthetic data pipeline in Isaac Sim to export image outputs and object bounding boxes directly into standard COCO JSON format for immediate consumption by training pipelines. Show how to configure an Omniverse Replicator writer to output COCO-compatible datasets. | pass→pass | 20,450 | 34,190 | +67% | 1 | 1 | 0% | 3,228 | 7,561 | +134% | 0 | 0 | — |
▸case-08 We want our simulated robot in Isaac Sim to stream its current joint angles to an external ROS 2 network on topic `/joint_states`. Instead of setting up a custom socket server script in Python, configure the built-in OmniGraph ROS 2 joint state publisher node graph. | fail→pass | 15,962 | 23,891 | +50% | 1 | 1 | 0% | 2,607 | 3,975 | +52% | 0 | 0 | — |
▸case-09 We need to stream synthetic RGB camera frames from an Isaac Sim sensor into ROS 2 on topic `/front_camera/image_raw`. Show how to create the OmniGraph node pipeline connecting a render product to the ROS 2 image publisher node. | fail→fail | 17,877 | 18,507 | +4% | 1 | 1 | 0% | 2,917 | 3,939 | +35% | 0 | 0 | — |
▸case-10 A robot chassis USD prim is suspended in mid-air because physics dynamics are disabled on its USD schema. We need a Python script to programmatically turn this USD prim into an active PhysX rigid body and assign explicit mass attributes. | pass→pass | 11,449 | 14,812 | +29% | 1 | 1 | 0% | 2,312 | 3,302 | +43% | 0 | 0 | — |
▸case-11 When running large-scale synthetic data generation jobs on a multi-GPU workstation with four NVIDIA RTX GPUs, we need to programmatically configure Isaac Sim settings to enable GPU physics and multi-GPU rendering pipeline execution. Provide Python code modifying the application settings. | pass→pass | 29,359 | 18,687 | -36% | 1 | 1 | 0% | 4,341 | 3,620 | -17% | 0 | 0 | — |
▸case-12 We want to package our custom synthetic data workflow tool as a formal Omniverse Kit extension. Provide the extension configuration file (`extension.toml`) defining package metadata and declaring core Isaac Sim extension dependencies. | fail→pass | 23,825 | 10,255 | -57% | 1 | 1 | 0% | 1,224 | 2,276 | +86% | 0 | 0 | — |
▸case-13 To train a monocular depth estimation network, we need distance-to-camera depth maps produced alongside color images in Isaac Sim Replicator. Rather than computing depth from stereoscopic disparity post-render, write a Replicator script attaching the native metric depth annotator. | pass→pass | 17,699 | 18,967 | +7% | 1 | 1 | 0% | 2,749 | 3,313 | +21% | 0 | 0 | — |
▸case-14 In a robotic pick-and-place benchmark, target objects must spawn at randomized 3D positions and rotations on a surface during every synthetic image generation step. Write a Replicator script that applies spatial translation and rotation distributions to object prims. | fail→fail | 17,664 | 17,860 | +1% | 1 | 1 | 0% | 3,314 | 3,638 | +10% | 0 | 0 | — |
▸case-15 A multi-joint robotic arm imported into Isaac Sim breaks apart into unconstrained free-floating rigid bodies during physics steps because articulation reduced coordinates are not initialized. Show how to apply the PhysX articulation root API to the robot's base prim in Python. | pass→pass | 15,551 | 18,299 | +18% | 1 | 1 | 0% | 2,325 | 3,768 | +62% | 0 | 0 | — |
▸case-16 For autonomous driving perception evaluation, we need synthetic sensor datasets formatted according to the KITTI benchmark structure (3D bounding boxes, calibration matrices, image files). Show how to configure an Omniverse Replicator writer for KITTI formatting. | pass→pass | 25,589 | 35,523 | +39% | 1 | 1 | 0% | 5,140 | 7,554 | +47% | 0 | 0 | — |
▸case-17 When running physics simulation in non-realtime speed, connected ROS 2 control nodes experience time drift. Write an OmniGraph Python script to publish simulation clock signals to the standard ROS 2 `/clock` topic. | pass→pass | 13,268 | 18,684 | +41% | 1 | 1 | 0% | 2,638 | 3,774 | +43% | 0 | 0 | — |
▸case-18 A detailed CAD mesh of an industrial engine block causes severe physics performance degradation when assigned a raw triangle mesh collider schema. Provide a Python script using USD Physics APIs to configure convex hull decomposition on the collision mesh. | pass→pass | 16,704 | 17,561 | +5% | 1 | 1 | 0% | 3,193 | 3,696 | +16% | 0 | 0 | — |
▸case-19 We need to programmatically assemble a clean USD stage lighting rig in Isaac Sim containing an ambient Dome Light and a directional Distant Light. Provide Python code using UsdLux APIs to build this lighting setup. | pass→pass | 15,881 | 17,569 | +11% | 1 | 1 | 0% | 3,176 | 3,795 | +19% | 0 | 0 | — |
▸case-20 To make synthetic camera data realistic for pose estimation algorithms, we need to inject Gaussian perturbation into camera extrinsic coordinates on each rendering step in Replicator. Write a Replicator script modifying camera transform parameters dynamically. | pass→pass | 15,395 | 25,971 | +69% | 1 | 1 | 0% | 2,881 | 4,382 | +52% | 0 | 0 | — |
▸case-21 We are developing a reinforcement learning control loop in Python where a simulated quadruped robot takes actions in Isaac Sim. Write Python code using `omni.isaac.core` to initialize the simulation world and step physics forward in the training loop. | pass→pass | 15,584 | 17,960 | +15% | 1 | 1 | 0% | 3,053 | 3,868 | +27% | 0 | 0 | — |
▸case-22 We are building a custom quadruped robot and need to write C++ firmware for an STM32F4 microcontroller to generate 20kHz PWM signals for direct BLDC motor gate drivers over CAN bus. Provide the C++ HAL code for Timer register setup. | pass→pass | 30,057 | 22,295 | -26% | 1 | 1 | 0% | 6,165 | 4,714 | -24% | 0 | 0 | — |
▸case-23 We are developing a web-based client application in React and WebGL using Three.js to display interactive 3D product models on e-commerce websites. Provide JavaScript code initializing a Three.js WebGLRenderer, Scene, and PerspectiveCamera. | pass→pass | 13,275 | 11,255 | -15% | 1 | 1 | 0% | 2,629 | 2,685 | +2% | 0 | 0 | — |
▸case-24 We are maintaining a legacy robotics simulation running Gazebo Classic 11 on ROS Noetic. We need a C++ plugin inheriting from nav_core::BaseGlobalPlanner to implement A* search on costmap_2d maps. Write the C++ header and source code for Gazebo Classic. | fail→pass | 32,415 | 27,654 | -15% | 1 | 1 | 0% | 7,012 | 6,615 | -6% | 0 | 0 | — |