東京工業大学 長谷川晶一研究室 : Hasegawa Shoichi Laboratory
日本語 | English

Pliant Motion  - Integration of Virtual Trajectory Control into LCP Based Physics Engines -

rep.png

We propose a realtime method to generate a virtual trajectory tracking any real trajectory for any articulated bodies with dynamics simulator based on iterative LCP solvers. Proposed method computes joint torques to track desired (real) trajectory using physics engines (1st step), then executes one step of simulation to update posture and velcity of the articulated body with PD control targeting to the desired trajectory (2nd step).

Introduction

Interactive applications such as video games require characters, which generate motions corresponding to user's interaction. Because the user's operation changes the motion trajectory, unexpected contact to objects may happen. The amount of change on a trajectory depends on not only the trajectory of motion but also internal tensions of skeletal muscles - co-contraction level, when a person put one's hand or a hand collide to an object. Reaching motion of human is supposed to be generated by spring damper characteristics of muscles dragging to the virtual trajectory. Human controls not only trajectories of motions but also spring-damper characteristics of muscles by changing co-contraction levels. Tracking of virtual trajectory generate realistic motions with unexpected contacts. However, it is not easy to make a virtual trajectory by hand.

Thus, we propose a realtime method to generate a virtual trajectory tracking any real trajectory for any articulated bodies with dynamics simulator based on iterative LCP solvers. Proposed method computes joint torques to track desired (real) trajectory using physics engines (1st step), then executes one step of simulation to update posture and velcity of the articulated body with PD control targeting to the desired trajectory (2nd step).

Because the computed torque from 1st step is almost enough to track desierd trajectory, the feedback coefficients for the PD control are not matter. Therefore, the generated trajectories have no difference between small and large feedback coefficients. Though proposed method do not compute virtual trajectory explicitly, virtual trajectory can find from the torques computed in 1st step by supposing that the torques are generated from the PD control targeting to the virtual trajectory. Proposed method mimics this motion generation method of human and easily generates realistic motion trajectories. Proposed method genearte motion trajectory in realtime with low computatioal amount using LCP solvers in physics engines.

Publications

Test

normal paragraph.