Abstract
Animations are a core component of video games. Animations typically require dedicated animators and are relatively inflexible, making it extremely difficult to animate a character without enforcing strict restrictions on the virtual world where it is placed. We propose a framework to explore the procedural generation of animations in arbitrary tridimensional virtual worlds, by using neuroevolution to create and evolve neural networks that output forces at a creature’s skeleton joints in order to produce motion and movement that are credible and physically coherent with the virtual world topology and the creature's state. Evaluation was done by selecting tasks for which our framework was able to generate animations that are different between each other but that achieve the same result proving that neuroevolution can offer different solutions to animation problems even when using different body topologies which were also able to achieve the same task using the same fitness method.