← ArchivePaper2023

Learning Robust and Scalable Motion Matching with Lipschitz Continuity and Sparse Mixture of Experts

Motion matching approach with Lipschitz-constrained networks and sparse mixture of experts for robust, scalable learned motion synthesis.

Open the paper ↗Trace in the lineage graph

Abstract

Motion matching(Büttner and Clavet [2015]; Clavet [2016]) has become a widely adopted technique for generating high-quality interactive animation systems in video games. However, its current implementations suffer from significant computational and memory resource overheads, limiting its scalability in the context of modern video game performance profiles. "Learned Motion Matching"[Holden et al. 2020] mitigated some of these challenges, however, whilst reducing memory requirements, it resulted in increases in performance costs. In this paper, we propose a novel method for learning motion matching that combines a Sparse Mixture of Experts model architecture and a Lipschitz-continuous latent space for representation of poses. This approach significantly reduces the computational complexity of the models, while simultaneously improving the compactness of the data that can be stored and the robustness of pose output. As a result, our method enables the efficient execution of motion matching that significantly outperforms other implementations for large character counts, by 8.5x times in CPU execution cost and at 80% of the memory requirements of "Learned Motion Matching", on contemporary video game hardware, thereby enhancing its practical applicability and scalability in the gaming industry.

How to read this

Category: Method: a learned motion matching architecture for games
Contributions: A learned motion matching method combining a Sparse Mixture of Experts model with a Lipschitz-continuous latent pose space
Reduced computational complexity and more compact stored data while improving robustness of pose output
Reported large efficiency gains over prior implementations for high character counts in CPU cost and memory
Context: Builds directly on Learned Motion Matching (which reduced memory but raised runtime cost) and on the original motion matching technique from games practice.Builds on: Learned Motion Matching
Correctness: Performance claims (e.g. large CPU speedups, reduced memory at high character counts) are stated relative to specific baselines and hardware profiles; treat the headline multipliers as conditional on that comparison setup, and motion quality versus the gains is the trade to watch.
Clarity: Game-systems oriented and accessible; a first pass conveys the SMoE-plus-Lipschitz idea, a second pass is needed for how Lipschitz continuity is enforced and measured.
How to read it: First pass for the motivation (scalability of motion matching) and the architectural recipe; second pass on the Lipschitz constraint and expert routing if you target large crowds at runtime.

Builds on

Learned Motion Matching 2020

Built upon by

Nothing yet.

Related work

Learned Motion Matching 2020 / SIGGRAPH
Example-based Motion Synthesis via Generative Motion Matching 2023 / SIGGRAPH
Real-Time Diverse Motion In-Betweening with Space-Time Control 2024 / MIG
Taming Diffusion Probabilistic Models for Character Control 2024 / SIGGRAPH

Keywords

motion matching sparse mixture of experts lipschitz continuity neural networks latent space pose autoencoder character animation learned motion matching encoder-decoder smoe lipschitz regularization stepper model projector model 6d rotation representation

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 →