Acquisition of Stand-up Motion by a Four-link Robot using Emergent Learning

摘要

The purpose of this research is to obtain a motion control method, which has adaptive and emergent ability. To develop a motion controller, we use RNN (Recurrent Neural Network) and GA (Genetic Algorithm). We applied this architecture to a 4-link robot for the task of learning stand-up motion. We also applied a learning method that utilizes mechanical assistance to the robot in order to improve a speed of learning. With decreasing degree of the assistance, the robot learns a suitable motion starting from an easy problem to a more difficult one. Simulation results show that the above approaches are effective.