We have described learning based approach for a robotic arm to grasp an object or clear the clutter kept in front of it. We have used AlexNet pre-trained ImageNet for our CNN (Convolutional neural network). We continuously feed the image to our Deep neural network and this deep neural network identifies the objects and gives out an output in the form of grasp angle and the coordinates of the object that is to be picked. These coordinates are fed to the robotic arm to do the action. Thus, implementing hand eye coordination. Our experimental results demonstrate that the robotic arm is able to grasp novel objects successfully. We got an accuracy of 70% for previously known objects and 64% accuracy for novel objects. © 2019 IEEE.