▸case-01 We are integrating a point-cloud based object detector with our manipulation pipeline. Can you create a grasp planning configuration that evaluates antipodal quality scores and provides the necessary pre-grasp and post-grasp gripper control actions? | pass→pass | 27,928 | 30,914 | +11% | 1 | 1 | 0% | 4,895 | 7,329 | +50% | 0 | 0 | — |
▸case-02 Our mobile manipulator needs to navigate across the warehouse floor from the charging station to the picking table using costmaps and Dijkstra path planning. Can you write the room-scale navigation configuration for this path planning task? | fail→fail | 24,040 | 17,384 | -28% | 1 | 1 | 0% | 3,968 | 3,627 | -9% | 0 | 0 | — |
▸case-03 We are setting up a new RGB-D camera and need to calculate the 3x3 intrinsic matrix and lens distortion coefficients using a checkerboard calibration target. Can you generate the intrinsic matrix optimization code for calibrating this sensor? | fail→fail | 17,459 | 22,930 | +31% | 1 | 1 | 0% | 3,738 | 4,121 | +10% | 0 | 0 | — |
▸case-04 We are flashing updated real-time EtherCAT motor controller firmware onto the joint actuators of our robotic arm. Can you provide the low-level flashing scripts and register bitwise configurations for updating the motor controller EEPROM? | fail→fail | 35,305 | 29,204 | -17% | 1 | 1 | 0% | 3,459 | 6,160 | +78% | 0 | 0 | — |
▸case-05 We are building an unorganized bin-picking pipeline where items are randomly oriented in 3D space. A developer suggested simplifying this by using 2D top-down bounding boxes and fixed z-heights. How should we structure a robust 6-DOF grasp pose estimation configuration that handles arbitrary 3D orientations from raw point cloud data? | pass→pass | 24,639 | 23,771 | -4% | 1 | 1 | 0% | 3,579 | 4,746 | +33% | 0 | 0 | — |
▸case-06 We need to evaluate candidate grasps on a complex 3D object mesh. A naive heuristic just selects surface normal pairs at random without checking physical stability. How do we configure antipodal grasp detection to ensure contact force closure under surface friction constraints? | pass→pass | 22,311 | 29,377 | +32% | 1 | 1 | 0% | 3,493 | 6,490 | +86% | 0 | 0 | — |
▸case-07 When generating 50 candidate grasps for a cluttered shelf scene, our controller currently selects the first kinematically reachable pose regardless of stability. How should we set up quantitative grasp quality metrics to rank candidates before execution? | fail→pass | 17,332 | 20,208 | +17% | 1 | 1 | 0% | 2,890 | 3,970 | +37% | 0 | 0 | — |
▸case-08 During picking execution, our end-effector collides with nearby items because it plunges straight into the grasp pose from oblique angles. How do we configure formal approach vectors to align the gripper before final entry? | pass→pass | 18,109 | 17,963 | -1% | 1 | 1 | 0% | 3,161 | 3,533 | +12% | 0 | 0 | — |
▸case-09 After closing the gripper on a part, lifting vertically immediately scrapes surrounding items in a deep bin. How do we specify custom retreat vectors in the manipulation plan to safely extract the object? | pass→pass | 17,523 | 15,866 | -9% | 1 | 1 | 0% | 2,870 | 3,215 | +12% | 0 | 0 | — |
▸case-10 Our parallel-jaw gripper attempts to navigate into tight spaces while wide open, colliding with adjacent items before reaching the target. How should we define pre-grasp standoff poses and jaw apertures to prevent premature contact? | pass→pass | 16,029 | 18,883 | +18% | 1 | 1 | 0% | 2,538 | 3,532 | +39% | 0 | 0 | — |
▸case-11 Our pick-and-place sequence moves the robot arm immediately after sending a closure command without verifying if the object was captured, causing frequent dropped items. How should we configure post-grasp state monitoring and execution verification? | pass→pass | 23,003 | 26,008 | +13% | 1 | 1 | 0% | 2,996 | 3,797 | +27% | 0 | 0 | — |
▸case-12 We generated top-scoring theoretical grasps for a part, but several of them place the physical gripper fingers inside a container wall. How do we implement collision-aware grasp filtering against environmental obstacle geometry? | pass→pass | 21,567 | 21,107 | -2% | 1 | 1 | 0% | 3,533 | 3,938 | +11% | 0 | 0 | — |
▸case-13 When manipulating a vessel like a mug, our planner treats all surface meshes equally and frequently attempts to grasp the open top rim instead of the handle. How can we integrate object affordance detection into candidate pose evaluation? | pass→pass | 19,252 | 22,302 | +16% | 1 | 1 | 0% | 2,973 | 4,311 | +45% | 0 | 0 | — |
▸case-14 Our cell repeatedly picks standard industrial fasteners, but recalculates heavy 6-DOF grasp candidates on every cycle. How do we set up a persistent grasp database to cache and retrieve verified grasp configurations by part identifier? | pass→pass | 20,929 | 25,659 | +23% | 1 | 1 | 0% | 3,694 | 5,575 | +51% | 0 | 0 | — |
▸case-15 We are writing `moveit-manipulation-planning.js` for an automated cell. An engineer wants to hardcode joint values for picking. How do we properly structure this script to integrate MoveIt grasp generation pipelines? | pass→fail | 134,229 | 24,138 | -82% | 1 | 1 | 0% | 3,508 | 4,889 | +39% | 0 | 0 | — |
▸case-16 We are implementing `object-detection-pipeline.js` to process vision sensor inputs for downstream picking. How should we format the output schema of this script to seamlessly feed pose and geometry data to the grasp planner? | fail→pass | 36,616 | 22,591 | -38% | 1 | 1 | 0% | 1,551 | 3,843 | +148% | 0 | 0 | — |
▸case-17 In `rl-robot-control.js`, our policy outputs continuous velocity vectors but fails to trigger discrete grasping effectively. How should grasp candidate quality and gripper trigger actions be exposed to the reinforcement learning controller? | pass→pass | 16,291 | 23,435 | +44% | 1 | 1 | 0% | 2,700 | 4,545 | +68% | 0 | 0 | — |
▸case-18 We are creating `hri-interface.js` for collaborative assembly. The human operator needs to visually inspect and confirm planned grasps before execution. How should candidate poses and affordance markers be structured for display? | pass→pass | 21,320 | 20,997 | -2% | 1 | 1 | 0% | 3,815 | 4,448 | +17% | 0 | 0 | — |
▸case-19 We need to generate candidate grasp poses directly from unorganized 3D point cloud clusters in a bin-picking setup where surface meshes are unavailable. Which tool framework should we integrate and how should sampling be configured? | fail→pass | 20,629 | 16,800 | -19% | 1 | 1 | 0% | 3,348 | 2,924 | -13% | 0 | 0 | — |
▸case-20 We want to run offline physics simulation and energy-based stability evaluations for candidate grasps on CAD models before physical deployment. Which tool framework should we integrate and how should quality scoring be configured? | fail→pass | 23,609 | 20,094 | -15% | 1 | 1 | 0% | 3,641 | 3,611 | -1% | 0 | 0 | — |
▸case-21 When attempting to grasp a thin sheet-metal part resting flat on a workbench, the lower finger of our parallel-jaw gripper impacts the tabletop. How do we configure approach clearance offsets and stroke parameters to ensure clean contact? | pass→pass | 18,734 | 21,008 | +12% | 1 | 1 | 0% | 2,817 | 3,623 | +29% | 0 | 0 | — |
▸case-22 Our arm uses a single pneumatic vacuum cup end-effector instead of finger jaws. How do we configure grasp pose generation and quality metrics tailored for suction cup surface attachment? | pass→fail | 22,776 | 25,983 | +14% | 1 | 1 | 0% | 3,299 | 4,777 | +45% | 0 | 0 | — |