[source] Siggraph
[year] 2002
Controllable, reactive human motion
combining dynamic simulation and human motion data.
use trajectory tracking to follow motion capture data and a balance controller to keep the character upright while modifying sequences from a small motion lib. to accomplish specified tasks.
react to forces computed from a physical collision model by changing stiffness and damping terms.
1. Introduction.
Motion capture-driven数据驱动, often used, but interaction affects believablilty –> modifying data
Physical model物理模型
combining 二者要混合了…
数据驱动，上希望得到动作细节；物理呢，来加点反应性动作
2.背景
围绕上述三个方面来的
3. Motion Capture-driven control
结合trajectory tracking 与 a process of selecting among and interpolating between sequences of human motion data.
PDs，但这里还没有说平衡的问题（反馈），在6中
4. Control for Hitting
修改动作数据，对修改后的数据跟踪
两个方式改变末端效应器:1.改变关节角度以到目标位置、旋转 2.time-scaling来调整速度
1. –> IK
2. –> 每段动作模拟下，根据原速度和新速度调整timestep,则角速度得到speed-up or speed-down,线速度自然也会跟着变
5. Reacting to Contact
During contact, the gains of the affected joints are reduced to allow the dynamics of the impact to influence the nmotion in a natural manner.
在tracking时，使用high gain,但当contact时会不自然(stiffness),更换k_r, b_r。
离线调，只调碰撞部位
调整模式
6. Balancing while tracking
control systems for balance that maintain the postion and velocity of the COM projected onto the ground plan
reduce the acceleration of the COM
two approaches: 1.a virtual actuator computes joint torques for balance and bombines them with the tracking torque in the lower extremeties 2. a second controller offsets the desired angle in the hips and ankles before computing torques
1: Virtaul actuator方法： 腿的关节上， OF_v = k_v (x_d – x) – b_v(\dot x) 再计算力矩，再转为局部坐标下的
2: : \lambda_{(h,a)} = k_{(h,a)}(x_d – x) – b_{(h,a)}(\dot x)