Rising up is a very common and important motion


Download 445 b.
Sana10.01.2019
Hajmi445 b.



Rising up is a very common and important motion

  • Rising up is a very common and important motion

    • Human / robot / avatar could fall and need stand up
    • reflects physical capability and style variation
  • Rarely addressed in computer animation



Rich variations

  • Rich variations

    • various lying postures
    • various environments
    • different characters (style, physical capability)
  • Complex motor skills

    • collision avoidance
    • balance maintenance
    • adaptation


Small database for typical rising motions

  • Small database for typical rising motions

  • Motion planning for large variations

  • Dynamics filtering for small variations



Most varieties appear at lying-to-squatting



Connects an arbitrary lying pose to database

  • Connects an arbitrary lying pose to database



Ensures physical plausibility

  • Ensures physical plausibility

  • Adapts to environments and characters



Composable controllers

  • Composable controllers

    • Faloutsos et al., SIGGRAPH 2001
  • Contact-rich motion control

    • Liu et al., SIGGRAPH 2010
  • Both focus on motion control of various types of motions

  • Not address the motion varieties

    • crucial for rising up motions


Hot topic in humanoid research

  • Hot topic in humanoid research

    • Morimoto and Doya, IROS’98
    • Fujiewara et al. IROS’03
    • Hirukawa et al., IJRR’05
    • Kanehiro et al., ICRA’07
  • Focus on robustness instead of varieties and flexibilities



Address analysis rather than generation of rising motions

  • Address analysis rather than generation of rising motions

    • McCoy and VanSant, Physical Therapy, 1993
    • Ford-Smith and VanSant, Physical Therapy, 1993


Motion Planning Problem



Rapidly-exploring random tree (RRT)



RRT-connect [Kuffner et al. 2000]



RRT-connect [Kuffner et al. 2000]



RRT-connect [Kuffner et al. 2000]



RRT-connect [Kuffner et al. 2000]



RRT-connect [Kuffner et al. 2000]



RRT-connect [Kuffner et al. 2000]



RRT-blossom [Kalisiak & van de Panne, 2006]

  • Blossom

    • add multiple samples
    • explore space more quickly


RRT-blossom





Connecting Posture Selection

  • Posture

  • Posture difference

  • Accelerating search by clustering the motion database



Motion Planning Strategies



RRT-blossom Modifications

  • RRT-blossom is originally proposed for lower-dimensional configuration space

  • To handle motion planning in high- dimensional posture space

    • plan global orientation and joint angle separately
  • Impose joint limit constraint and avoid collision in the blossom operation



Dynamics Filtering

  • Track a planned motion using velocity-driven control [Tsai et al., TVCG 2010]

  • Balance by virtual actuator control [Pratt et al.]



Dynamics Filtering (cont.)

  • In some cases, our controller may not be able to track from squatting to standing

    • connect to a nearest rising motion in the database
    • fine since less variations from squatting to standing


Results

  • Our database only has14 motions of rising up on flat ground (CMU MOCAP database)

  • Rising up from random initial postures

  • Rising up with an initial and a key posture

  • Rising up in various environments

  • Motion retargeting of rising up

























Quality evaluation by human subjects

  • score range from 10 (best) to 1 (worst)

  • 27 males and 13 females aged 19 to 60



Conclusion

  • Simple and effective approach

    • Small database + motion planning + dynamics filtering
  • Generate rising up motions with varieties

    • various lying postures and environments
    • physically plausible
  • Efficient motion planning strategy

    • Loose-to-strict spatiotemporally local refinement strategy




Limited database + dynamics simulation

  • Limited database + dynamics simulation

    • 14 rising motions
  • Motion planning

    • increases motion varieties
    • avoids collisions
  • Dynamics filtering



Rapidly-exploring random tree (RRT)



RRT-connect [Kuffner et al. 2000]



Approach Overview

  • Stage I: connecting posture selection

    • Given Pinit, find a closest Pcon
  • Stage II: motion planning

  • Stage III: dynamics filtering

    • Tracking the planned motion





Do'stlaringiz bilan baham:


Ma'lumotlar bazasi mualliflik huquqi bilan himoyalangan ©fayllar.org 2017
ma'muriyatiga murojaat qiling