Optimization and generalization of whole-body cooperative motion of a humanoid robot using a genetic algorithm and neural networks

When a humanoid robot pushes an object with its force, it is essential to adequately control its posture so as to maximize the surplus torque ratio for all joints. For such purpose, this study proposes a method to find an optimal posture of a humanoid robot using a genetic algorithm (GA) in such a way that the surplus torque ratio for all joints is maximized in such cases. In this method, structure analysis is conducted by ANSYS for computing the force and moment exerted on the legs. As a result of successive maximization processes, an optimized posture having a large surplus joint torque at a desired pushing force is finally found. These optimization processes are conducted at various pushing force, then a number of optimized postures are found. Then, to apply such posture control to real time control, generalization process are conducted by a multilayer perceptron(MLP) trained by the obtained optimized postures and the pushing forces. To show the effectiveness of the proposed method, pushing motion of a 24-DOF humanoid robot is considered. The simulation results show that the proposed method can be used for real time control based on the optimized postures at various pushing forces.