← ArchivePaper2020

Local Motion Phases for Learning Multi-Contact Character Movements

Sebastian Starke, Yiwei Zhao, Taku Komura, Kazi Zaman

SIGGRAPHAcademic242 cites4 descendantsMotion Synthesis

Local phase representation per body part enabling neural networks to learn multi-contact character movements without global phase ambiguity.

Open the paper ↗Trace in the lineage graph

Abstract

Training a bipedal character to play basketball and interact with objects, or a quadruped character to move in various locomotion modes, are difficult tasks due to the fast and complex contacts happening during the motion. In this paper, we propose a novel framework to learn fast and dynamic character interactions that involve multiple contacts between the body and an object, another character and the environment, from a rich, unstructured motion capture database. We use one-on-one basketball play and character interactions with the environment as examples. To achieve this task, we propose a novel feature called local motion phase, that can help neural networks to learn asynchronous movements of each bone and its interaction with external objects such as a ball or an environment. We also propose a novel generative scheme to reproduce a wide variation of movements from abstract control signals given by a gamepad, which can be useful for changing the style of the motion under the same context. Our scheme is useful for animating contact-rich, complex interactions for real-time applications such as computer games.

How to read this

Category: Method: a motion representation for learning multi-contact character movement
Contributions: Local motion phase, a per-bone phase feature that helps networks learn asynchronous limb movements and contacts with external objects.
A framework that learns fast, contact-rich interactions (body-object, body-character, body-environment) from rich unstructured motion capture.
A generative scheme that reproduces wide motion variation, including style changes under the same context, from gamepad control signals.
Context: Builds on Starke et al.'s Neural State Machine for Character-Scene Interactions, addressing the global-phase ambiguity that arises in fast multi-contact motion.Builds on: Neural State Machine for Character-Scene Interactions
Correctness: Validated on examples such as one-on-one basketball play and environment interaction; as a data-driven real-time method, fidelity depends on the motion database and the local-phase labeling, and very novel contact configurations are the natural stress test.
Clarity: Accessible in concept, dense in detail; a first pass conveys why local phases beat a global phase, a second pass is needed for the feature extraction and network.
How to read it: Concentrate on the definition and extraction of local motion phase and why it resolves asynchronous contacts; a second pass on the generative control scheme is worth it for real-time applications.

Builds on

Neural State Machine for Character-Scene Interactions 2019

Built upon by

Related work

Neural State Machine for Character-Scene Interactions 2019 / SIGGRAPH Asia
A Deep Learning Framework for Character Motion Synthesis and Editing 2016 / SIGGRAPH
Phase-Functioned Neural Networks for Character Control 2017 / SIGGRAPH
Mode-Adaptive Neural Networks for Quadruped Motion Control 2018 / SIGGRAPH

Keywords

local motion phase motion capture neural character controller mixture of experts gating network character control multi-contact interaction basketball generative control model adversarial training phase functioned neural networks evolutionary strategy autoregressive motion synthesis quadruped locomotion contact labels inverse kinematics gamepad control

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 →