A Study of Applying an Artificial Neuromolecular System to Control a Quadruped Chopstick Robot

Jong-Chen Chen and Chung-Chian Hsu


Evolutionary Learning, Robot, Neural Networks, Memetic Computing


Memetic Computing (MC) has been proposed as a new paradigm in dealing with computationally intractable engineering optimization problems. It integrates the merits of population-based evolutionary learning and cultural evolutionary learning. The artificial neuromolecular system (ANM) that we constructed earlier [1][2] is an MC model comprised of two types of neurons. The first type of neurons possesses significant intraneuronal dynamics that captures biological structure-function relationship, an important feature for local/individual refinements. The second type of neurons acts on the repertoire of the neurons with internal dynamics to perform coherent functions, a form of population-based evolutionary learning. In this study, the ANM system was tested with a quadruped robot (named Miky) comprised of 16 disposable chopsticks, with each foot being controlled by an actuator (motor). The ANM system served to integrate input signals in space and time into a sequence of output signals, which in turn determine the movement of the Miky’s feet (through actuators). The task was to teach Miky either to walk straight or to make a turn. An inclinometer was installed at the center-top of the Miky’s skeleton for providing feedback to the ANM system during the course of learning. Our experimental result showed that Miky was capable of learning in a continued manner.

Important Links:

Go Back