最佳化搏擊力道角色控制

Abstract

本論文欲合成全身動作，做出武術攻擊動作並且保持身體平衡。我們在物理模擬環境下控制人類骨架的近似模型，使用最佳化演算法(Optimization algorithm) 找到合適的動作。只要給定目標函數(Objective function) 以及輸入初始狀態，就可以找到一組近似最佳解的動作。我們使用的目標函數考慮了拳頭要攻擊的位置以及拳頭的末速度，欲擊中攻擊目標並給予目標最大的衝量，擊中物體時會有反作用力作用到人類骨架上影響全身的平衡，因此我們也使用重心位置計算能量函數來保持平衡。最後我們得到的結果是虛擬角色可以在物理模擬中出拳攻擊指定的位置，同時以雙腳與身體其他部位保持自身的平衡。

The goal of this thesis is the synthesis of full-body motions to achieve desired martial arts attack actions and keep the human body balanced. We control an approximated model of the human skeleton in the physical simulation environment and use optimization algorithm to find an approximate optimal solution of desired motion. Given objective functions and the initial state of the skeleton, our method can find a motion sequence that minimizes the energy function values (local minimal). We use positions of limbs and the speed of hands in optimizing objective functions. We give a position of the target we want a character to attack. If its right hand does not pass the position at a specified time, the penalty increases. The objective function of balance will increase the penalty value when the center of mass of full-body is away from the target at desired position. Finally, we achieve the goal that a human-character can punch a target at specified position, while keeping its full-body balance.