Skip to content

← ArchivePaper2024

Universal Humanoid Motion Representations for Physics-Based Control

Zhengyi Luo, Jinkun Cao, Josh Merel, Alexander Winkler, Jing Huang, Kris Kitani, Weipeng Xu

ICCVAcademic172 citesMotion SynthesisRetargeting

PULSE learns a universal motion imitator from large unstructured mocap then builds a comprehensive physics-based motion representation.

Abstract

We present a universal motion representation that encompasses a comprehensive range of motor skills for physics-based humanoid control. Due to the high dimensionality of humanoids and the inherent difficulties in reinforcement learning, prior methods have focused on learning skill embeddings for a narrow range of movement styles (e.g. locomotion, game characters) from specialized motion datasets. This limited scope hampers their applicability in complex tasks. We close this gap by significantly increasing the coverage of our motion representation space. To achieve this, we first learn a motion imitator that can imitate all of human motion from a large, unstructured motion dataset. We then create our motion representation by distilling skills directly from the imitator. This is achieved by using an encoder-decoder structure with a variational information bottleneck. Additionally, we jointly learn a prior conditioned on proprioception (humanoid's own pose and velocities) to improve model expressiveness and sampling efficiency for downstream tasks. By sampling from the prior, we can generate long, stable, and diverse human motions. Using this latent space for hierarchical RL, we show that our policies solve tasks using human-like behavior. We demonstrate the effectiveness of our motion representation by solving generative tasks (e.g.

How to read this

Category
Method / representation: universal motion latent space for physics-based control
Contributions
  • Learns a motion imitator that can reproduce all of human motion from a large, unstructured mocap dataset
  • Distills skills from the imitator into a universal motion representation via an encoder-decoder with a variational information bottleneck
  • Jointly learns a proprioception-conditioned prior enabling long, stable, diverse motion generation and hierarchical RL on downstream tasks
Context
Builds directly on Luo et al.'s Perpetual Humanoid Control, broadening prior narrow, style-specific skill embeddings into a comprehensive universal representation (PULSE).Builds on: Perpetual Humanoid Control for Real-time Simulated Avatars
Correctness
Assumes a strong full-coverage imitator can be distilled into a reusable latent without losing skill breadth; demonstrated for downstream hierarchical RL control, though the abstract is truncated, so check the paper for which tasks, coverage limits, and failure modes apply.
Clarity
Conceptually dense (VAE bottleneck, distillation, hierarchical RL); a first pass conveys the imitate-then-distill idea, later passes are needed for the latent-space and prior details.
How to read it
Focus on the two-stage imitate-then-distill design and the role of the variational bottleneck and proprioceptive prior; a second pass on the representation pays off if reusing it as a downstream control prior.

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 →