▸case-01 Can you provide a configuration script for a 3D object detection network such as PointPillars or VoxelNet using 3D point cloud data, along with a wrapper node that outputs standard ROS vision_msgs for object detection pipelines? | pass→pass | 58,197 | 38,496 | -34% | 1 | 1 | 0% | 4,792 | 7,387 | +54% | 0 | 0 | — |
▸case-02 We are setting up visual SLAM for our indoor mobile robot using an RGB-D sensor. Can you write a ROS node that constructs a 2D occupancy grid map using Cartographer? | fail→fail | 43,103 | 29,728 | -31% | 1 | 1 | 0% | 3,988 | 5,505 | +38% | 0 | 0 | — |
▸case-03 Our robotic arm needs to plan collision-free joint trajectories to pick up an object from a table. Can you generate a MoveIt 2 C++ node that calls the OMPL trajectory planner and executes the trajectory via FollowJointTrajectory? | fail→fail | 24,210 | 19,048 | -21% | 1 | 1 | 0% | 4,206 | 4,297 | +2% | 0 | 0 | — |
▸case-04 We want to perform pairwise 3D point cloud registration to stitch multiple LiDAR sweeps together using Iterative Closest Point (ICP). Can you provide a Python script for ICP registration using Open3D? | fail→fail | 19,039 | 26,765 | +41% | 1 | 1 | 0% | 2,812 | 5,104 | +82% | 0 | 0 | — |
▸case-05 We need a Python script to convert an exported ONNX object detection model to an optimized engine file specifically targeted for NVIDIA Jetson devices with FP16 precision. Most tutorials default to standard CPU ONNX runtime conversions. Write the script using the TensorRT Python API. | pass→pass | 19,519 | 28,317 | +45% | 1 | 1 | 0% | 4,097 | 5,348 | +31% | 0 | 0 | — |
▸case-06 We are deploying a real-time object detection model on an edge device using ONNX Runtime Python API. Engineers often forget to set GPU execution providers and end up running CPU execution. Write an ONNX Runtime inference script that properly configures CUDA execution. | pass→pass | 14,916 | 27,233 | +83% | 1 | 1 | 0% | 2,978 | 5,005 | +68% | 0 | 0 | — |
▸case-07 We are building an instance segmentation pipeline for a robotic bin-picking application requiring pixel-accurate object masks and polygon prediction. Instead of simple bounding box detectors, select the appropriate standard PyTorch framework for instance segmentation and write a Python inference snippet using its `DefaultPredictor` configuration. | pass→pass | 12,950 | 14,543 | +12% | 1 | 1 | 0% | 2,556 | 3,243 | +27% | 0 | 0 | — |
▸case-08 We need a ROS 2 Python node that subscribes to an uncompressed camera topic (`/camera/image_raw`), runs a real-time YOLO detector, and outputs standardized ROS detection messages. Instead of creating custom ROS message types, output standard messages from the vision_msgs package. | pass→pass | 22,405 | 28,174 | +26% | 1 | 1 | 0% | 3,738 | 5,683 | +52% | 0 | 0 | — |
▸case-09 We are implementing a multi-object tracking system for an autonomous mobile robot monitoring pedestrian flow. Simple frame-by-frame detection loses object IDs during occlusions. Select an appropriate multi-object tracking algorithm suitable for high-performance detection-based tracking and write a Python snippet that updates track IDs across frames. | pass→pass | 23,096 | 29,120 | +26% | 1 | 1 | 0% | 3,389 | 5,432 | +60% | 0 | 0 | — |
▸case-10 We want to train a custom YOLO model (v8) for real-time detection on a mobile robot. Developers often write complex manual training loops in PyTorch. Provide the standard Ultralytics Python API code to load a pretrained lightweight model (`yolov8n.pt`), train on a custom dataset YAML file, and export the trained model directly to ONNX format. | pass→pass | 7,266 | 15,216 | +109% | 1 | 1 | 0% | 1,345 | 2,256 | +68% | 0 | 0 | — |
▸case-11 We have an RGB camera paired with an aligned depth camera (`/camera/aligned_depth_to_color/image_raw`). We want to extract 3D spatial coordinates (X, Y, Z in meters) for detected 2D object bounding boxes. Write a Python function that uses pixel bounding box centroids, depth values, and camera intrinsic parameters (fx, fy, cx, cy) to compute 3D camera-frame coordinates. | pass→fail | 21,254 | 27,953 | +32% | 1 | 1 | 0% | 3,569 | 4,216 | +18% | 0 | 0 | — |
▸case-12 We are setting up a multi-camera object detection fusion node for an autonomous vehicle with front and rear cameras. Instead of running separate independent perception topics without merging, write a Python script that takes overlapping bounding box detections from two camera nodes in world coordinates and merges duplicate detections using distance-based non-maximum suppression (NMS). | pass→pass | 26,247 | 34,492 | +31% | 1 | 1 | 0% | 4,293 | 6,744 | +57% | 0 | 0 | — |
▸case-13 We need to optimize a semantic segmentation model for embedded deployment on NVIDIA Jetson. Write a Python script using PyTorch and CUDA events to profile model inference latency and compute frames per second (FPS) over a benchmark loop. | pass→pass | 21,252 | 27,516 | +29% | 1 | 1 | 0% | 3,385 | 4,969 | +47% | 0 | 0 | — |
▸case-14 We want to build a synthetic data pipeline for generating labeled 2D bounding box ground truth images for training object detectors in robotics. Provide a Python script using OpenCV to composite object foreground cutouts onto randomized background images, outputting annotations matching the COCO JSON dataset spec. | pass→pass | 31,421 | 39,824 | +27% | 1 | 1 | 0% | 6,455 | 7,599 | +18% | 0 | 0 | — |
▸case-15 We are deploying a real-time vision pipeline on ROS 2 with high bandwidth camera streams. Engineers often write unoptimized callbacks causing high CPU load. Write a C++ ROS 2 node using image_transport and cv_bridge to convert `sensor_msgs::msg::Image` to OpenCV `cv::Mat` for neural network preprocessing. | pass→pass | 27,623 | 27,692 | +0% | 1 | 1 | 0% | 4,547 | 5,067 | +11% | 0 | 0 | — |
▸case-16 We need a Python script using ONNX Runtime to execute a dual-output neural network that outputs both object bounding boxes and a semantic segmentation map. Provide the inference loop that retrieves dynamic output names from the session and evaluates model outputs. | pass→pass | 23,648 | 32,451 | +37% | 1 | 1 | 0% | 3,918 | 7,248 | +85% | 0 | 0 | — |
▸case-17 For a mobile manipulator, we need a 3D bounding box detector that ingests 3D PointCloud2 messages. Provide a Python script configuring PointPillars inference to output 3D bounding box parameters including 3D center positions (x, y, z), dimensions (dx, dy, dz), and orientation yaw angle. | pass→pass | 46,532 | 35,632 | -23% | 1 | 1 | 0% | 5,005 | 6,788 | +36% | 0 | 0 | — |
▸case-18 We are building a SORT (Simple Online and Realtime Tracking) tracker node in Python to assign persistent IDs to YOLO bounding box outputs. Write a Python snippet that uses the Hungarian Algorithm via scipy to associate frame detections with existing track predictions. | pass→pass | 18,702 | 29,238 | +56% | 1 | 1 | 0% | 3,998 | 4,776 | +19% | 0 | 0 | — |
▸case-19 We need to set up semantic segmentation preprocessing for an autonomous indoor robot navigating clutter. Write a Python script using torchvision or OpenCV to normalize RGB camera images using ImageNet mean and standard deviation, and convert model class output logits into a discrete 2D class segmentation mask. | pass→pass | 25,167 | 27,814 | +11% | 1 | 1 | 0% | 4,089 | 4,974 | +22% | 0 | 0 | — |
▸case-20 We want to implement depth-assisted object detection filtering on an RGB-D camera to eliminate false-positive object detections outside a target operating distance of 3.0 meters. Write a Python function that accepts 2D detection bounding boxes and a depth map, calculates median depth within each box region, and filters out distant detections. | pass→pass | 25,005 | 24,265 | -3% | 1 | 1 | 0% | 3,287 | 4,441 | +35% | 0 | 0 | — |
▸case-21 We are configuring TensorRT optimization for INT8 quantized object detection models on Jetson devices. Write a TensorRT Python script that defines an `IInt8EntropyCalibrator2` class to calibrate an ONNX detection model using sample calibration images. | pass→pass | 21,322 | 20,330 | -5% | 1 | 1 | 0% | 4,583 | 4,428 | -3% | 0 | 0 | — |
▸case-22 We need a ROS 2 vision pipeline node that performs real-time object detection and broadcasts dynamic transform frames for each detected object relative to `camera_color_optical_frame`. Write a Python ROS 2 node using `tf2_ros.TransformBroadcaster` to broadcast detected object transforms. | pass→pass | 29,137 | 31,175 | +7% | 1 | 1 | 0% | 6,235 | 6,039 | -3% | 0 | 0 | — |
▸case-23 We are optimizing a model deployment using TensorRT dynamic shape profiles on Jetson. When dynamic input resolution is needed (ranging from min 320x320 to max 1280x1280), static engine builds fail. Write the TensorRT Python API snippet that creates an IOptimizationProfile and configures min, optimal, and max dynamic dimensions for input tensors. | pass→pass | 13,158 | 15,280 | +16% | 1 | 1 | 0% | 1,666 | 2,357 | +41% | 0 | 0 | — |