Applications of Evolutionary Computation to
Quadrupedal Animal Animation
James E. Murphy, B.Sc. (Hons.)
Submitted to University College Dublin
for the degree of Ph.D. in the
College of Engineering, Mathematical and Physical Sciences
March 2011
Supervisors:
Dr. Michael O'Neill (site)
Dr. Hamish Carr (site)
Examiners: Prof. Simon Lucas, Prof. Anthony Brabazon and Dr. Michela Bertolotto
Thesis:
Link to electronic copy of thesis here.
Videos:
Links to video list: .mov and .mp4.
Abstract:
As humans are familiar with animal movement, a realistic animal animation must imitate this motion. This thesis explores how observations of natural evolution and evolutionary computation can be used to produce realistic quadrupedal animal animations, focusing on the grammar-based Genetic Programming method of Grammatical Evolution (GE).
A cross-discipline review of animation systems, biological knowledge and natural computing techniques applicable to animal animation is presented. Focusing on the horse, the construction of both a kinematic and physics-based model is described. The origins and representations of the data used to construct and animate these models are also discussed.
GE is applied to animation problems for the first time. Animating physics-based models is complex and a GE motion optimisation system successfully generates realistic, stable motions. Additionally, simple grammars with little domain knowledge generate novel movement.
For herd scenes, a GE-based system generates models of varying morphology and automatically optimises motion data for them. GE also evolves motion adjuster functions that dynamically modify limb movement based on the model's velocity. These functions are used in a real-time controllable kinematic animation system, in which a horse model moves with an accurate gait and executes gait transitions when necessary.
Overall, the use of GE and natural world observations is found to facilitate the generation of realistic quadrupedal animal animations.
Submitted to University College Dublin
for the degree of Ph.D. in the
College of Engineering, Mathematical and Physical Sciences
March 2011
Supervisors:
Dr. Michael O'Neill (site)
Dr. Hamish Carr (site)
Examiners: Prof. Simon Lucas, Prof. Anthony Brabazon and Dr. Michela Bertolotto
Thesis:
Link to electronic copy of thesis here.
Videos:
Links to video list: .mov and .mp4.
Abstract:
As humans are familiar with animal movement, a realistic animal animation must imitate this motion. This thesis explores how observations of natural evolution and evolutionary computation can be used to produce realistic quadrupedal animal animations, focusing on the grammar-based Genetic Programming method of Grammatical Evolution (GE).
A cross-discipline review of animation systems, biological knowledge and natural computing techniques applicable to animal animation is presented. Focusing on the horse, the construction of both a kinematic and physics-based model is described. The origins and representations of the data used to construct and animate these models are also discussed.
GE is applied to animation problems for the first time. Animating physics-based models is complex and a GE motion optimisation system successfully generates realistic, stable motions. Additionally, simple grammars with little domain knowledge generate novel movement.
For herd scenes, a GE-based system generates models of varying morphology and automatically optimises motion data for them. GE also evolves motion adjuster functions that dynamically modify limb movement based on the model's velocity. These functions are used in a real-time controllable kinematic animation system, in which a horse model moves with an accurate gait and executes gait transitions when necessary.
Overall, the use of GE and natural world observations is found to facilitate the generation of realistic quadrupedal animal animations.