Optimization of Artificial Central Pattern Generators with Evolutionary Algorithms

In contrast to classical engineering approaches for the generation of movements in robots or prostheses, approaches to this subject inspired by neurophysiological circuits are in advance. One of the key structures of interest in this area is the Central Pattern Generator (CPG) which has been identified to be the source of movement generation in mammals. This neural circuit is capable of generating cyclic muscle activation patterns completely independent from the brain as shown by Brown in the early 20th century [1, 2]. In the past years this knowledge was adapted to the challenges in movement generation and control in robotics, for example for passive dynamic walkers by P. Manoonpong [3]. In this paper the results of a CPG implementation and its optimization to model the human movement generation are presented, which were aiming at a design to be as biologically realistic as possible. Due to the fact, that biological models are the fundamentals of the simulated CPG, a large number of parameters needs to be set, which results in a very complex configuration process to get an optimized CPG behavior. Based on previous works [4–7] an Evolutionary Algorithm was employed to optimize the rhythm-generation of the CPG. For this optimization the tool GLEAMKIT was used [8]. Electromyographic (EMG) data recorded from a human male during walking on a treadmill have been utilized for the fitness function as well as for evaluation purposes.