Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Construction of physically based animats with genetic algorithms

: Vicente San, G.
: Bisler, A.

San Sebastián, 2005, 73 pp.
San Sebastián, Univ. de Navarra, Dipl.-Arb., 2005
Fraunhofer IGD ()
Artificial Life; autonomous agent; genetic algorithm

The presented project intends to imitate the locomotion of animals by constructing physically based animats (artificial animals). The main objective is to study man-made systems that exhibit characteristic behaviors of natural living systems. The program autonomously creates animats and optimizes them by using genetic algorithms. In order to model the animat, it is constructed by means of the physics engine ODE (Open Dynamics Engine). The body is made up of solid elements which are connected by joints. Locomotion is achieved by applying forces to the joints.
Instead of manually constructing and optimizing such an animat, a genetic algorithm is used. The information needed to construct an animat are encoded in a genome. Starting with a generation of randomly constructed animats, the algorithm evaluates the fitness of each animat. Therefore, the phenotype (the fully functional animat) is constructed using the information from its genome. During a simulation run the performance - how fast the animat can move - is measured. The better an animat performs, the higher its chance to pass its genome to the next generation. To achieve variations the copied genomes are mutated.
The project can be divided into three main areas: designing the structure of information that defines the animat both physically and with regards to movement, the immersion of the animat in a 3-dimensional environment, where it can move and interact, and the graphical representation of the simulation results. The results were visualized using OpenSG. For the genetic algorithm the GAlib and for the physics engine the ODE library had been used.