← ArchivePaper2018

Mode-Adaptive Neural Networks for Quadruped Motion Control

He Zhang, Sebastian Starke, Taku Komura, Jun Saito

SIGGRAPHAcademic376 cites10 descendantsMotion Synthesis

Mode-adaptive neural network with learned gating for quadruped locomotion control handling diverse gaits and transitions.

Open the paper ↗Trace in the lineage graph

Abstract

Quadruped motion includes a wide variation of gaits such as walk, pace, trot and canter, and actions such as jumping, sitting, turning and idling. Applying existing data-driven character control frameworks to such data requires a significant amount of data preprocessing such as motion labeling and alignment. In this paper, we propose a novel neural network architecture called Mode-Adaptive Neural Networks for controlling quadruped characters. The system is composed of the motion prediction network and the gating network. At each frame, the motion prediction network computes the character state in the current frame given the state in the previous frame and the user-provided control signals. The gating network dynamically updates the weights of the motion prediction network by selecting and blending what we call the expert weights, each of which specializes in a particular movement. Due to the increased flexibility, the system can learn consistent expert weights across a wide range of non-periodic/periodic actions, from unstructured motion capture data, in an end-to-end fashion. In addition, the users are released from performing complex labeling of phases in different gaits. We show that this architecture is suitable for encoding the multi-modality of quadruped locomotion and synthesizing responsive motion in real-time.

How to read this

Category: Method: a data-driven neural controller for character locomotion
Contributions: Introduces Mode-Adaptive Neural Networks for quadruped motion control
Pairs a motion-prediction network with a gating network that blends specialized expert weights per frame
Learns from unstructured mocap end-to-end, removing manual gait labeling and phase alignment
Context: Builds on phase-based neural character control (Phase-Functioned Neural Networks), generalizing the periodic phase mechanism into a learned gating over experts for non-periodic and periodic quadruped actions.Builds on: Phase-Functioned Neural Networks for Character Control
Correctness: Demonstrated on quadruped locomotion across varied gaits and actions learned from mocap; a reader should note that the gating/expert capacity and training-data coverage bound which motions and transitions can be reproduced.
Clarity: Moderately technical; a first pass conveys the gating-of-experts idea, and a second pass is needed for the network formulation and training setup.
How to read it: First pass for the gating-network-blends-experts concept versus a fixed phase function; do a second pass on the architecture and loss if you intend to implement or adapt it to other morphologies.

Builds on

Phase-Functioned Neural Networks for Character Control 2017

Built upon by

Neural State Machine for Character-Scene Interactions 2019

Related work

Neural State Machine for Character-Scene Interactions 2019 / SIGGRAPH Asia
Near-Optimal Character Animation with Continuous Control 2007 / SIGGRAPH
Local Motion Phases for Learning Multi-Contact Character Movements 2020 / SIGGRAPH
Character Motion Synthesis by Topology Coordinates 2009 / CGF

Keywords

mode-adaptive neural networks quadruped motion control locomotion synthesis gating network motion prediction network neural network architecture mixture of experts character animation animal motion capture real-time 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 →