A Two-Stage Trajectory Planning Method for Online Robotic Quality Measurement

Robotic optical measurement systems have been widely used in various fields. However, facing the demand for small-scale customized manufacturing style, the traditional offline measurement planning methods are difficult to satisfy the high-efficiency requirement. To address this challenge, this article proposes the use of external sensors to collect environmental data alongside the original measurement system. Furthermore, a two-stage online planning framework is proposed to generate measurement trajectories for unknown target objects. In Stage 1, the feature recognition stage, a point cloud segmentation network considering multineighborhood features is proposed to recognize point clouds in small target regions (i.e., boundaries of holes) from large background areas. In Stage 2, the trajectory generation stage, a novel generative adversarial network (GAN) is utilized for end-to-end robotic trajectory generation directly using point clouds of inspection features. In addition, a variety loss function is proposed to consider the unique attributes of the inverse kinematics of industrial robots and improve the accuracy of the generated trajectory. To verify the effectiveness of the proposed framework, a robotic measurement system consisting of a UR5 robot, a line laser scanner, and a depth sensor was built. The results indicate that, compared with the benchmark methods, the proposed method demonstrates several advantages. First, it achieves optimal segmentation accuracy and planning efficiency. Second, it improves planning accuracy and efficiency in simulation scenarios by 50% and 61%, respectively. Third, it enhances planning efficiency and detection accuracy in real-world scenarios by 76.7% and 50.0%, respectively.