Design of the Indoor WiFi Cloud Positioning System Based on GANand Its Application to Service Robots

To overcome the poor indoor positioning accuracy and expensive equipment of service robots, an indoor WiFi cloud positioning system based on generative adversarial networks (GANs) is proposed. Firstly, the overall architecture of the cloud positioning system is designed in detail, and the method of regressive wireless positioning is proposed based on GAN. The network parameters are pre-trained by GAN, which summarizes and extracts the positioning values from the massive data sources automatically. Therefore, the low accuracy of the traditional algorithms from hand-engineering is avoided. Secondly, the global backpropagation optimization of the network parameters is carried out by the fully connected neural network. Finally, Kalman filter is applied to filtering jump points to solve the positioning dispersion problem caused by turbulent wireless signals, so that the real-time positioning is smoother and more accurate. A lot of experiment and application results show that compared with the traditional positioning algorithms, the wireless positioning method based on GAN for service robots can obtain a much higher accuracy, and the average error of positioning is reduced within 0.29 m, which meets the demand of service robots for positioning.