Application of single neuron model to motion planning and control of under-actuated robot by MDP framework

Motion planning of robots based on the Markov decision process (MDP) is now one of the most important topics of robotics research. In motion planning of robots by the MDP, an efficient way to divide robot motion is essential because discrete expressions of states and transitions are used in the calculation. In preparing the discrete expressions of robot motion, it is important that the motion follows the MDP, second, that the time and the amount of memory for the planning calculation are reduced, and that the maneuverability and smoothness of the robot motion are not lost. This paper proposes a new discrete expression of robot motion that can meet these requirements considering the effect of robot dynamics to robot kinematics. The method uses a continuous-time continuous-variable neuron model, which has the stimulation and adaptation properties. The performance of the proposed method was examined by simulation of the motion planning of an under actuated underwater robot and under actuated manipulator with a free joint. The results show the high reliability and flexibility of the method.