← ArchivePaper2021
Neural Animation Layering for Synthesizing Martial Arts Movements
Neural animation layering system enabling real-time synthesis of complex martial arts movements by composing motion layers with learned networks.
Abstract
Interactively synthesizing novel combinations and variations of character movements from different motion skills is a key problem in computer animation. In this paper, we propose a deep learning framework to produce a large variety of martial arts movements in a controllable manner from raw motion capture data. Our method imitates animation layering using neural networks with the aim to overcome typical challenges when mixing, blending and editing movements from unaligned motion sources. The framework can synthesize novel movements from given reference motions and simple user controls, and generate unseen sequences of locomotion, punching, kicking, avoiding and combinations thereof, but also reconstruct signature motions of different fighters, as well as close-character interactions such as clinching and carrying by learning the spatial joint relationships. To achieve this goal, we adopt a modular framework which is composed of the motion generator and a set of different control modules. The motion generator functions as a motion manifold that projects novel mixed/edited trajectories to natural full-body motions, and synthesizes realistic transitions between different motions.
How to read this
- Category
- Method: neural animation layering for motion synthesis
- Contributions
- A deep learning framework that imitates animation layering to mix, blend, and edit movements from unaligned motion-capture sources in a controllable way
- A modular design of a motion generator (acting as a motion manifold) plus separate control modules that synthesizes novel locomotion, punching, kicking, avoiding, and combinations with realistic transitions
- Reconstruction of fighter-specific signature motions and close-character interactions such as clinching and carrying by learning spatial joint relationships
- Context
- Builds on Starke et al.'s Local Motion Phases for multi-contact movements and echoes the layered, scriptable-actor idea of Perlin's Improv, applying learned manifolds to martial-arts motion composition.Builds on: Local Motion Phases for Learning Multi-Contact Character Movements · Improv: A System for Scripting Interactive Actors in Virtual Worlds
- Correctness
- Assumes that a learned motion manifold can project mixed or edited trajectories back to natural full-body motion; demonstrated on martial-arts mocap, so readers should note generalization beyond the trained motion domain and interaction types is not established here.
- Clarity
- Accessible at the systems level; a first pass conveys the layering analogy and modular structure, a second pass clarifies the generator and control-module networks.
- How to read it
- Read pass one for the layering-as-neural-modules concept and the manifold idea; a second pass on the control modules and phase handling pays off if you want to build interactive real-time synthesis.
Builds on
Built upon by
Nothing yet.
Related work
- Dog Code: Human to Quadruped Embodiment Using Shared Codebooks 2024 / MIG
- DeepPhase: Periodic Autoencoders for Learning Motion Phase Manifolds 2022 / SIGGRAPH
- A Deep Learning Framework for Character Motion Synthesis and Editing 2016 / SIGGRAPH
- Automated Extraction and Parameterization of Motions in Large Data Sets 2004 / 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 →