Motion Modeling for Online Locomotion Synthesis Taesoo Kwon

Motion Modeling for Online Locomotion Synthesis Taesoo Kwon and Sung Yong Shin KAIST

Outline • Motivation • Related work • Overview • Motion analysis • Motion synthesis • Conclusions • Future Work

Motivation • Real-time locomotion synthesis • Motion rearrangement : realism • Motion blending : efficiency and controllability • Hybrid approach – Locomotive motion generation [PSS 02, PSS 04] – Rhythmic motion synthesis [KPS 03] • Premise: motion labeling

Related Work • Motion Segmentation [Bindiganavale & Badler, 1998; Fod et al. , 2002; Kim et al. , 2003] • Motion Classification [Arikan et al. , 2003; Kovar & Gleicher, 2004; Forbes & Fiume 2005; Mueller & Roeder 2005] • Motion Labeling for blending [Kim et al. , 2003]

Overview example motions motion analysis hierarchical motion transition graph motion specifications motion synthesis desired motion

Motion Analysis • Issues – Motion segmentation – Motion classification – Graph construction • Biomechanical observations – [Per 92, Win 90]

Biomechanical Observations • Center of mass trajectory walk right foot left foot transition run

Motion segmentation • Criteria for motion segmentation – Simple enough for intuitive parameterization – Long enough to contain motion semantics – An important motion feature should not be split Split at every COM peak

Motion Classification • String encoding – • Pros – avoid troublesome time-warping – more robust than numerical computation

Motion Classification • Footstep patterns (a) S (b) R (c) L (d) D (e) F

Motion Classification • String Encoding (ideal case)

Motion Classification • String Encoding (ideal case) R D L

Motion Classification • String Encoding (ideal case) F R F

Motion Classification • String Encoding (ideal case) R D L F

Motion Classification • String Encoding (ideal case)

Refinement • False peak – Concatenate two motion segments • Missing peak – Divide a motion segment into two

Graph Construction

Graph Construction

Motion Analysis Results • O(n) – 2 Ghz PC (AMD 64, 2 GB memory) – For 7. 4 min locomotion, about 10 seconds • Movie

Motion Synthesis … LDR RDL LDRF …

Motion Synthesis • Motion specification • Motion parameter

Motion Sythesis • How to calculate – Two half cycles in cyclic motion – • Regression analysis on

Motion Synthesis • Motion blending : [PSS 04][KG 03][ACP 02] • Motion stitching : [GSKJ 03] • Motion retargeting : [SLSG 01][KGS 02]

Motion Synthesis Result • 1000+ frames per second • Movie – Path following – Online synthesis

Conclusion • Motion labeling based on string encodings • Hierarchical motion transition graph

Future work • Footstep-driven motions such as dancing and boxing