Virtual fluid environment on behavior ability for artificial creature

An environment plays an important role in behaviors acquisition for artificial creatures. Thus, the environment must obey the physical laws. In this paper, it is examined how the behavior differences appear when the artificial creature autonomously behaves in some fluid environments. We construct the approximate virtual fluid environment with low computing costs to simulate the behavior acquisition for artificial creatures. Also we propose a simulation method for artificial creatures in consideration of effects from the virtual fluid environment based on physics modeling. As a result of simulation, we verify that it is possible for the creature to acquire adaptive behaviors in different environments. After evolution, the creature behaves autonomously by leveraging effectively fluid forces in each virtual environment. Introduction Many computer simulations have been performed for studying acquiring behaviors, evolution, and learning methodologies on virtual artificial life creatures in the field of Artificial Life (ALife) and evolving robotics. Artificial fish swims automatically by learning its behavior controller (Tu and Terzopoulos, 1994). This study makes it easy to create fish animation. A flock simulation approach is developed based on a distributed behavioral model without setting the orbit of each bird (Reynolds, 1987). This approach makes it easy to create flock animation. The virtual creature is able to acquire its morphology and behavior by an evolutionary methodology based on the creature’s competition (Sims, 1994a)(Sims, 1994b). Many studies for behavior acquisition are based on Sims’ studies. Sims’ model is applied to the virtual catapult creatures to evolute (Chaumont et al., 2007). This creature could throw its parts of body as far as possible. The relation for co-evolution of virtual creatures is observed by fighting each other in Sims’ virtual environment (Miconi, 2008). In these days, there are many simulations for artificial creature using the physical calculating engine. It enables these creatures to obey physics law easily. ”SnakeLike Robot” acquires adaptive locomotion on the ground using it (Tanev et al., 2005). This model is robust for obstacles. In these studies, the experimental environment is set as an ideal environment in a computer simulation space because they considered that the methodology of evolving and learning behavior in an ideal environment is more important than acquisition of the similar behavior in realistic environment. Therefore, the influenced force from the fluid environments to the artificial creature is not precisely analyzed. Instead, the implemented force adopts the simple calculation methods for reducing the computing time. On the other hand, in a field of numerical fluid dynamics, many fluid simulations are based on a finite element method and a particle method. The moving particle semi-implicit method makes it easy to create animation on the water surface (Koshizuka et al., 1998). A virtual anomalocaris model swims in the virtual two-dimensional water environment using the particle method (Usami, 2007). And an artificial creature behaves based on a rule method considering the fluid effect (Lentine et al., 2010). The finite element method and the particle method give accurate results. However, they consume much computational time. Therefore, it is unsuitable for a real-time simulation to acquire appropriate behaviors in the virtual fluid environment. However, we consider that the virtual environment needs to acquire a more natural policy of adaptive behaviors. In this paper, it is examined how the behavior differences appear when the artificial creature autonomously behaves in the different fluid environments. At first, we construct an approximate virtual fluid environment which enable us to do the behavior acquisition simulation with low computing costs. This environment is constructed by setting physics parameters such as the fluid density and drag coefficients. And we propose a simulation method for the artificial creature in consideration of the fluid environment. The artificial creature imitating a flat fish is modeled by connecting rigid bodies. This creature can behave by moving its bodies. In order to control bodies and learn the behaviors, an artificial neural network (ANN) is implemented with the creature. Genetic algorithm (GA) is applied to the ANN by its evolution. We experiment to examine how the artificial creature can acquire adaptive behaviors in some fluid environments. As a result of simulation, we verify that it is possible to ac-