Implementation of an improved parallel metaheuristic on GPU applied to humanoid robot simulation

Generally, bio-inspired techniques require significant computational resources. However, due to their complexity and the computing power required for their execution, they have long been neglected. Nevertheless and recently, parallel resolution techniques exploiting the graphics processing units (GPUs) are increasingly used. These specialized processors are being widely adopted for the purpose of taking care of problems requiring very massive computations, particularly in the field of Bioinformatics and artificial life. This paper presents a new methodology for the implementation of a GPU processor of a bio-inspired technique realizing the evolution of walking behavior for a simulated humanoid robot. Our technique enables efficient mapping between the explored research space and the hierarchy´s own GPU memory. In conclusion, we have obtained significant results since the acceleration rate, compared to a sequential approach, varies between 4.68 and 11, 36.