Overview of Picking Accuracy

Hardware issues: The camera itself may be faulty or damaged, such as problems with the lens, sensor, or other components.

Environmental changes: Changes in environmental factors such as lighting conditions, temperature, humidity, etc., may affect the performance of the camera, resulting in a decrease in absolute accuracy, such as camera temperature drift.

Mechanical deformation: Deformation or loosening of the mechanical structure of the camera or its mounting frame may result in instability of the camera’s position, which can affect absolute accuracy.

Point Z-value repeatability (1σ): One standard deviation of 100 Z-value measurements of the same point. Point Z-value repeatability is used to evaluate the accuracy of a single pixel in the Z direction.

Z-region repeatability (1σ): One standard deviation of the Z-mean differences between the two regions measured 100 times in the depth map. Z-region repeatability (1σ) is used to evaluate the accuracy of the fit plane in the Z-direction.

Working distance: The larger the working distance, the faster the repeatability decreases.

Hardware issues: The camera itself may be faulty or damaged, such as problems with the lens, sensor, or other components.

Environmental changes: Changes in environmental factors such as lighting conditions, temperature, humidity, etc., may affect the performance of the camera, resulting in a decrease in repeatability, such as camera temperature drift.

Mechanical deformation: Deformation or loosening of the mechanical structure of the camera or its mounting frame may result in instability of the camera’s position, which may affect repeatability.

Stress deformation, loss, or aging of internal components.

Changes in the external environment, such as temperature, humidity, pressure, etc.

Warm-up drift refers to the point cloud drift caused by the temperature change of the 3D camera during the cold-start process.

Environmental thermal drift refers to the point cloud drift caused by changes in ambient temperature and humidity.

Preheating and warm-up: In order to reduce the occurrence of inaccurate picking, the camera needs to be preheated and warmed up when hand-eye calibration, camera intrinsic parameter calibration, intrinsic parameter correction, or calibration of other compensation parameters. You can warm up the camera by one of the following methods:

Keep the temperature, humidity and pressure of the environment in which the application operates relatively stable.

Deploy a "System Drift Self-Correction" system: It can not only solve the problem of camera temperature drift, but also ensure the reliability and operational stability of the 3D vision system.

Robot repeatability refers to the degree to which the robot’s position fluctuates when it repeatedly reaches a point in space.

Robot absolute accuracy refers to the deviation from the target position when the robot reaches a point in space. Measuring the difference between the set linear distance and the actual movement distance is a rough way to determine the absolute accuracy of the robot.

Robot zero position is lost.

TCP accuracy error.

The robot is not mounted securely.

Wrong robot poses were used in the hand-eye calibration, resulting in large error with the calibration results or calibration failure.

The vision system outputs incorrect vision results to the robot.

The vision system outputs incorrect waypoints to the robot, resulting in picking inaccuracy, picking failure, or collision with environmental objects.

Qualify of the 2D images captured by the camera: If the captured 2D image is overexposed or underexposed, it directly affects the inference result of deep learning. This issue is usually solved by adjusting the 2D exposure parameters. In the presence of strong ambient light, shading measures should be taken to improve image quality.

Quality of deep learning model: The quality of the acquired 2D image and the trained deep learning model directly impact deep learning inference. Clear and high-quality images for model training and representative datasets improve the deep learning result.

Deep learning parameter settings: Appropriate setting of deep learning parameters improves the accuracy of instance segmentation. By fine-tuning these parameters, you can adjust the model performance in different scenarios to achieve the best deep learning inference result. Periodically review and update parameter settings to accommodate possible system changes.

Point cloud quality: If the quality of the point cloud captured by the camera is poor, it directly affects the quality of the created point cloud model and scene point cloud, thus reducing the accuracy of 3D matching.

Accuracy of point cloud mode and pick points: Surface model and edge model provide different matching accuracy. Surface model provides higher matching accuracy, but the recognition speed is slower when the surface model is used. When making the model, you can add pick points by dragging or teaching. Pick points added by teaching are more accurate. Therefore, in scenarios where high accuracy is required, workpiece orientations are relatively consistent, and robot TCP errors are difficult to evaluate, it is recommended to add pick points by teaching to improve accuracy.

Setting of the matching algorithm: Inappropriate matching algorithm setting may have a negative impact on matching results and repeatability. To ensure accurate matching, correctly select and configure the matching algorithm, especially in complex scenarios and when workpiece types need to be switched. Adjust algorithm parameters in time to meet actual requirements and improve matching effect and system stability.

Lack of comprehensive consideration: Sometimes, when designing pose adjustment strategies, there may be a lack of thorough consideration for various positional changes and situational differences that a workpiece might undergo. Insufficient comprehensive consideration can lead to inadequate adaptability to specific scenarios.

Delay in updating to environmental changes: Work environments may change due to layout modifications, equipment replacements, or alterations in how workpieces are positioned. If pose adjustment strategies are not promptly updated to reflect these changes, it can render the strategy ineffective and unable to accurately adapt to new working scenarios.

Lack of flexibility: Some pose adjustment strategies may be designed too rigidly, lacking the flexibility to accommodate the diversity in workpiece positions. Insufficient flexibility can result in deviations in situations that were not adequately considered.

Point cloud processing: The lighting environment, material characteristics of the workpiece, and the reflective properties of surrounding objects significantly impact the effectiveness of the workpiece point cloud. To enhance the robustness of point cloud processing, comprehensively consider and adapt to different lighting conditions and object surface properties to reduce repeatability errors.

Hardware influence: The influence of temperature on camera hardware will affect the intrinsic parameters and extrinsic parameters of the camera and the accuracy of pose transformation. Therefore, keeping temperature changes and adjusting extrinsic and extrinsic parameters in time help ensure system stability.

Vision recognition method selection: The stability of the system is directly affected by whether to use instance segmentation, the matching algorithm selection, matching parameter settings, and pick point settings. Choose a vision recognition method that is suitable for specific scenarios and optimize matching parameter settings. This helps the system adapt more effectively to various working conditions.

Logical judgment and error prevention: The combination of logical judgment and error prevention is also crucial. Poorly designed logical judgments or error prevention functions may cause the system to handle abnormal situations improperly, thereby affecting overall stability.