基于GAN的服务机器人室內WiFi云定位系统设计与实现

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

针对服务机器人室内定位精度差、设备昂贵等问题,提出了基于生成对抗网络(GAN)的服务机器人室内WiFi云定位系统.首先,详细设计了云定位系统的整体架构,提出了基于GAN的回归无线定位方法,采用GAN对网络参数进行预训练,从海量的数据源中归纳、提取定位价值特征,避免了传统算法因为需要人为指定特征而导致的低精确度问题;其次,通过全连接神经网络反向传播方法对网络参数进行全局优化;最后,针对无线信号波动所造成的定位离散问题,利用卡尔曼滤波滤除跳点,从而使实时定位更加平缓准确.大量实验及应用结果表明,基于GAN的定位算法与传统定位算法相比,大幅度提高了服务机器人无线定位的精度,平均定位误差优于0.29 m,满足服务机器人对定位的需求.

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.