چكيده به لاتين
Walking robots present a new group in robots family that have gained lots of attentions due to their different abilities from other robots. During the last decades, several advanced research laboratory robots have been under investigations. In this thesis, control of quadruped robots was investigated. Walking stability, constraints satisfaction and forward movement was guaranteed by solving constraint optimization problem. In order to find the best answer, the optimization process was applied at every sampling time. Model Predictive Control (MPC) is selected for optimization at every sampling time. This control method has many advantages such as considering future events. Fulfillment of constraints in robot’s forward movement and its future performance is discussed in this method. The prediction model sends an estimation of the system model to the controller and based on that, the system future behavior was discussed. Considering receding horizon for analyzing system behavior leads to finding better control inputs. However, time consuming calculation and high solving time for each control cycle is one of the model predictive control problems. For practical purposes, non-real-time methods cannot be applied. To benefit from the advantages of the model predictive control and at the same time reducing its calculation time , hybrid methods must be applied. In this thesis, combining model predictive control method with projection recurrent neural network as the optimizer, solving time was decreased significantly. However, incorporating constraints to this optimizer is a challenging problem that has been circumvented by some proposed methods in this thesis. Finally, after proving stability for the presented method, it was applied to a laboratory quadruped robot. The results show drastic reduction in calculation time, which leads to real-time feasibility of the proposed method.
