Identification of Motion and Prediction of Time Series for Mobile Robot Sensors based ANFIS and Neural Networks Using LSTM

Without employing a mathematical model, combination neural networks (NNs) and fuzzy logic control have been identified as the primary techniques for enhancing the performance of coupled nonlinear robot systems. Two-wheeled mobile robot system kinematic model identification algorithm is presented in this paper using Adaptive Network Based Fuzzy Inference System (ANFIS) and NNs. Depend on a Modified Elman recurrent NN (MERNN) with a back propagation algorithm, the structure of the dynamic NNs will be analyzed that use to learn the neural modeling approach. The performance of the system can be enhanced by first training a NNs model offline, and then transferring that model into a process in which adaptive learning is carried out using ANFIS. According to the simulation findings, the suggested modelling can provide superior control performance in terms of reduced tracking error and precision in tracking trajectories between the mobile robot’s intended output and its actual actions (orientation and position) and the model of output neural. Furthermore, a set of rules for a fuzzy controller can be constructed without in-depth information about the system. A precise model is not necessary when using the ANFIS control design approach. This paper also make a comparison between LSTM, phased LSTM, and stacked LSTM architectures for time series prediction. The experimental data is motion sensor data collected during actual drives by a mobile robot. The LSTM stacking model superior the other two models, although all three models offer excellent predictions for the near future. Lastly, the actual and predicted trajectories of the robot are compared.