Séminaire IRIT-UT1 - Nour El Houda BENALIA

le 21 mars 2014

ME303 (plan)

Nour El Houda Benalia - Labo. LESIA (U. Biskra - Algérie)

Parallélisation d'algorithme de Vie Artificielle sur GPU

Bio-inspired methods generally need an interesting amount of computational resources.  However, due to their complexity and the long execution time involved with, it prohibits its use in many domains. Recently, parallel methods computation using graphic processing units have attracting much research interests.

Nowadays, many algorithms are rewritten and redesigned for modern graphics cards (GPU) which have a SIMD massive parallelism and this for a significantly lower cost. The type of interesting problems in our case is some cases in artificial life, especially those interested in the movement of artificial creatures.

 Researchers in biomechanics, robotics, and computer science work to understand human natural motion in order to reproduce it in other forms. The aim of humanoid robotic researchers is to obtain robots that can imitate the human behaviours to collaborate, in the best way, with humans. An obvious problem confronting humanoid robotics is the generation of stable and efficient gaits in a reasonable time.  In order to address this problem, alternative, biologically inspired control methods have been proposed, which do not require the specification of reference trajectories.

The objective of the first case of study is to propose a model that accelerates our method already proposed, in which is the combination of an evolutionary algorithm and a recurrent neural network that composes the brain of our robot.

The main challenge of legged robots is the issue of creating controllers for them. The problem is difficult because of the number of degrees of freedom in each leg and because of changes in the body center of mass and momentum. Generally, the controllers designed suffer from several problems for several reasons.

In order that the controller of the humanoid robot can emerge toward better solutions, he must take a great time. To overcome this problem, the obvious idea is to accelerate the evolutionary process by immersing the evolutionary algorithm used in GPUs. Our purpose is to propose a technique to accelerate the process of the controller in GPUs. Results are compared to the serial algorithm by studying the effect of a number of parameters.

Mis à jour le 18 avril 2014