Object recognition and strong grip detection are essential for intelligent robotic manipulation systems to function. They are especially important in factories and other places where the lighting is often complex and constantly changing. Shadows, reflections, glare and low light conditions can cause failure in traditional computer vision and simple deep learning-based methods, resulting in false object detection, false pose estimation and false robot grasp. So, the purpose of this project is to develop an object piece recognition and grip detection system from the deep learning. This is done by combining an improved YOLO Net with a feature fusion and attention module and the pose is obtained by using ROS based robotic controlThe proposed system consists of a YOLO based object detection network in the first part which employs improved feature fusion to integrate features of different sizes to be able to detect small and complex workpieces. The network has an embedded attention mechanism to highlight the features of the object and reduce noise from a non-uniformly lit environment. This leads to a more accurate and robust detection. Once the target object is found in the picture, the hand and eye calibration can be performed to find the exact change from camera frame to the robot hand frame. This allows us to find the exact location of the object in the world coordinate frame. Accurate pose information about the object may be very important to avoid mistakes in positioning while the robotic arm moves towards, grasps, and drops off the target object, which can lead to failures in grasping.The robot modelling module defines the poses of the manipulator such as kinematic information, dynamic parameters, joint limits, degree of freedom, etc. This ensures that the trajectory planning can be done. The components are integrated in ROS environment. The control module receives the detection results and pose information. The robot can grasp and place things smoothly and collisionlessly in real time.3.4. Experimental results The experimental results indicate that the proposed system can effectively solve the problems of unbalanced illumination and environmental interference. The recognition accuracy of the model is 92.2% and the overall average success rate for grasping is 93.75%. The experimental results confirm the feasibility, robustness and operational efficiency of the proposed system in actual industrial automation applications. The system is easily adaptable to different objects, lighting conditions, different levels of robotic arm working, and cooperation with other robots..
KUMAR et al. (Fri,) studied this question.