The current development of a robotic arm solution for the manufacturing industry requires performing pick-and-place operations for work pieces varying in size, shape, and color across different stages of manufacturing processes. It aims to reduce or eliminate the human error and human intervention in order to save manpower costs and enhance safety at the workplace. Machine learning has become more and more prominent for object recognition in these pick-and-place applications with the aid of imaging devices and advances in the image processing hardware. One of the key tasks in object recognition is feature extraction and object classification based on convolutional neural network (CNN) models, which are generally computationally intensive. In this paper, an intelligent object detection and picking system based on MobileNet is developed and integrated into an educational six-axis robotic arm, which requires less computation resources. An experimental test is conducted on six-axis robotic arm called Niryo One to train the model and identify three objects with difference shapes and colors. It is shown by the confusion matrix that the MobileNet model achieves an accuracy of 91%, a dramatic improvement compared to 65% of the Niryo One’s original sequential model. The statistical study also shows the MobileNet can achieve a higher precision with more clustered spread of accuracy.