Near-optimal Character Animation with Continuous Control
[source] siggraph
[year] 2007
对人物的交互式控制,在低维度上计算得到至目的地的近似最优动作。
关注的是fluidity and variability, synthesis of a kinematic controller that blends precaptured motion clips to achieve complex multi-objectives in real-time.
两部分: 1.motion model stitches together captured motion clips 2. a controller selects sequences of clips
Motion Model:
每个动作有一个in ,一个out,用以融合,每个阶段中都是单足与地面接触,以消除滑步
Control:
描述:X = (C,x,z,\theta, u,v \dot u, \dot v , \hat G, \hat \tao, \hat \omega)
f(X,C’) = X’ 为 X经片段C’时到状态X’,C为X原属片段
代价: A. cost to each state: c_s: S –> R
B. cost of transition: c_t : SxS—> R
用这两个代价构成来刻画一些任务,文中给出四种
Policies
要使总代价最小
1.贪心
2.最优
3近似最优
以基函数近似V函数,形成LP,解得组合参数
有几篇参考文献,计划略读下