← ArchivePaper2002
Motion Graphs
Builds a graph of transitions inside a mocap corpus so arbitrary streams of motion can be synthesized from clips.
Abstract
This paper introduces motion graphs, a method for synthesizing realistic and controllable character motion from a corpus of motion capture data. A directed graph is automatically constructed in which edges hold pieces of original motion data plus automatically generated transitions, and nodes serve as choice points where clips can be seamlessly connected. Candidate transitions are detected using a point-cloud distance metric over windows of frames, blended with linear and spherical linear interpolation, and the graph is pruned using strongly connected components to guarantee well-connected, label-consistent motion. New motion is generated by searching the graph with a branch and bound algorithm for graph walks that satisfy user constraints, demonstrated on the problem of synthesizing locomotion along arbitrary user-sketched paths.
How to read this
- Category
- Method: data-driven motion synthesis from a mocap corpus
- Contributions
- Motion graphs, an automatically constructed directed graph whose edges hold original clips plus generated transitions and whose nodes are connection choice points
- A point-cloud distance metric over frame windows to detect candidate transitions, blended with linear and spherical-linear interpolation, with pruning via strongly connected components
- A branch-and-bound graph search that synthesizes constrained motion, demonstrated on locomotion along user-sketched paths
- Context
- Builds on Rose et al.'s Verbs and Adverbs multidimensional motion interpolation, moving from blending parameterized clips toward graph-based reassembly of an entire capture corpus.Builds on: Verbs and Adverbs: Multidimensional Motion Interpolation
- Correctness
- Demonstrated on synthesizing locomotion along arbitrary user paths; reader caveat is that output quality is bounded by the captured corpus and by the transition metric and blending, so coverage and naturalness depend on the data and graph connectivity.
- Clarity
- Accessible; a first pass conveys the graph idea, a second pass clarifies the distance metric, pruning, and search.
- How to read it
- Focus on graph construction (transition detection plus connectivity pruning) and the search formulation; a second pass on the distance metric and branch-and-bound is worth it if you build a synthesis system.
Builds on
Built upon by
- Motion Synthesis from Annotations 2003
- Automated Extraction and Parameterization of Motions in Large Data Sets 2004
- Near-Optimal Character Animation with Continuous Control 2007
- Tiling Motion Patches 2012
- Motion Grammars for Character Animation 2016
- Motion Matching and The Road to Next-Gen Animation 2016
Related work
- DeepPhase: Periodic Autoencoders for Learning Motion Phase Manifolds 2022 / SIGGRAPH
- Character Controllers Using Motion VAEs 2020 / SIGGRAPH
- Motion Grammars for Character Animation 2016 / CGF
- A Deep Learning Framework for Character Motion Synthesis and Editing 2016 / SIGGRAPH
Keywords
This page summarises the entry and links to its original source. The archive never hosts or redistributes the publication itself.Show it in the full archive list →