Abstract
With the digitalization of industry, the need for reliable and interpretable robotic systems for object manipulation is growing. This paper examines an integrated ROS-based system that provides a complete cycle from visual object detection to its grasping and movement. The goal of the study is to develop a unified architecture that combines determinism, robustness to external influences, and practical applicability without the need for additional training of neural network models. The proposed approach incorporates classical computer vision methods, 3D scene reconstruction, 6D object pose estimation, analytical grasp planning, and collision-free motion planning in MoveIt. To improve system robustness, methods for temporal stabilization of estimates and cross-modal data verification were implemented, and a pipeline was developed that enables the selection of grasping strategies, execution of the «pick-lift-place» sequence, and recovery through re-detection and trajectory replanning. The practical significance of the work lies in the possibility of implementing the system in industrial logistics, warehouse automation, and service robotics tasks without significant reconfiguration to new conditions.