Integrating Convolutional Neural Network and Deep-Q-Network for the Performance Enhancement of Internet of Robotic Things

The Internet of Robotic Things (IoRT) plays an important role in various applications that help to make effective communication circumstances. However, the robots require the proper features about particular scenarios to make a clear decision. In addition, the system’s security, robustness and flexibility are difficult to maintain while analyzing various robotic application-related scenarios. The difficulties are addressed using integrated techniques such as Convolution Neural Networks (CNN) and Deep Q Networks (DQN) to improve the overall robotic performance, such as object identification and abnormal behaviour prediction. Initially, various scenarios of visual information are collected with the help of the camera. The gathered information is processed using a convolution kernel and pooling layer for downsampling and extracting the features. The derived hierarchical features are utilized to observe the scenario and identify the objects. The extracted features are fed into the DQN approach, which utilizes the rewards in the reinforcement learning process to improve decision-making efficiency. The decisions generated are used to fine-tune robotic performance in different applications. Then, the system’s efficiency is evaluated in various real-time application scenarios in which the IoRT system attains high robustness, security, flexibility, and reliability.