Skip to content

← ArchivePaper2023

Composite Motion Learning with Task Control

Pei Xu, Xiumin Shang, Victor Zordan, Ioannis Karamouzas

SIGGRAPHAcademic57 citesMotion Synthesis

GAN-based decoupled multi-discriminator framework learns composite body-part motions from multiple references without manual annotation.

Abstract

We present a deep learning method for composite and task-driven motion control for physically simulated characters. In contrast to existing data-driven approaches using reinforcement learning that imitate full-body motions, we learn decoupled motions for specific body parts from multiple reference motions simultaneously and directly by leveraging the use of multiple discriminators in a GAN-like setup. In this process, there is no need of any manual work to produce composite reference motions for learning. Instead, the control policy explores by itself how the composite motions can be combined automatically. We further account for multiple task-specific rewards and train a single, multi-objective control policy. To this end, we propose a novel framework for multi-objective learning that adaptively balances the learning of disparate motions from multiple sources and multiple goal-directed control objectives. In addition, as composite motions are typically augmentations of simpler behaviors, we introduce a sample-efficient method for training composite control policies in an incremental manner, where we reuse a pre-trained policy as the meta policy and train a cooperative policy that adapts the meta one for new composite tasks.

How to read this

Category
Method: composite, task-driven motion control for physics-based characters
Contributions
  • Learns decoupled per-body-part motions from multiple reference clips simultaneously using a multi-discriminator GAN-like setup, with no manual composite references
  • Proposes a multi-objective learning framework that adaptively balances disparate motion sources and goal-directed control objectives in one policy
  • Adds a sample-efficient incremental scheme that reuses a pre-trained policy as a meta policy and trains a cooperative policy for composite behaviors
Context
Built on the adversarial motion prior approach, extending AMP (Peng et al. 2021) from full-body imitation to decoupled, composable body-part control.Builds on: AMP: Adversarial Motion Priors for Stylized Physics-Based Character Control
Correctness
The control policy is left to discover how to combine motions automatically, which avoids annotation but means composition emerges from training rather than being guaranteed; demonstrated on physically simulated characters with multiple tasks.
Clarity
Moderately technical; a first pass conveys the multi-discriminator idea, a second pass is needed for the multi-objective balancing and incremental meta-policy scheme.
How to read it
Focus on how multiple discriminators map to body parts and how objectives are balanced; a second pass is worthwhile for the incremental, sample-efficient training and what 'cooperative policy' means.

Builds on

Built upon by

Nothing yet.

Related work

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 →