Research On Gait Driven Quadruped Locomotion Using Low-dimensional Physical Model

Character animation is one of the core subjects in the domain of computer animation.The main objects of character animation are to create motions efficiently and effectively for characters with constraints from different properties, environments, and motion objects. Both bipedal and quadruped characters are two major reseach objects of character animation. For bipedal characters, large amounts of data have been collected with widely used motion capture equipments. Therefore, motion synthesis and motion recognition research is able to be improved rapidly. However, in contrast to large diversity and wide variety of quadruped characters 一 mainly refers to four-legged terrestrial animals —, only few types of them could adapt existing motion captrure equipments. It caused the motion data deficient of quadruped animation, and hindered its researches on motion editing and synthesis. On the other hand, compare with bipedal characters, quadruped characters reveal more gait patterns, richer foot constraints, and higher structural balancing, this makes locomotion skills with physical model becoming an available method to gain quadruped motion data.Therefore, a quadruped locomotion method is proposed, which using low-dimensional physical model abstracted from quadruped kinematic model, and is driven by dynamical four-legged gait constraints. This method can generate customized quadruped locomotion rapidly, with directly input of gait graphs, speed,and few interactions.The major contribution of this dissertation includes:(i) A low-dimensional physical model for quadruped locomotion.A low-dimensional physical model for quadruped locomotion is proposed: First, the concepts of "mass point" and "rigid constraint" is introduced to describe "joint" and"skeleton" in character animation, and the corresponding physical solver are also given;Second, torso, limbs, head and tail are separately abstracted, with models of line springs and torsion springs; Finally, based on knowledge of physiological and physical degree of freedom, a queue of locomotion constraints are constructed for correcting locomotion results from physical model. Using this model, quadruped locomotion with visual reality could be generated rapidly with few interactions, and motion styles could be modified by tuning parameters.(ii) A dynamic gait constraint model for low-dimensional locomotion.A dynamic gait constraint model is proposed, which can plan gait patterns and parameters dynamically by the instant speed, and estimate the speed with gait graph input. After that, according to the gait parameters, speed, and the energy optimization principles in biological motions, an associational model with time, terrain, gait, and leg joint rotation is created, for real-time planning of terrain adaptive swing leg animation.Meanwhile, steering strategies for head, torso and feet are established, in order to generate feet and body constraints and drive physical model turning to the assigned direction.(iii) A fast locomotion system adapt to diverse groups of quadrupeds.On the basis of gait constraints model and low-dimensional physical model,researches for improving reusability of data and parameters are conducted to assist users processing fast locomotion with diversity styles. Those researches mainly including: automatic joint-matching and semantic recognition; morphologic similarity based multi-layer motion data retargetting, and interactive stylized locomotion by operating semantic style tags. Finally, all works above are integrated into a quadruped locomotion system, which can automatically import batches of character, rapidly generate objective locomotion, and support stylization with operations of semantic tags.