Skip to content

← ArchivePaper2024

PDP: Physics-Based Character Animation via Diffusion Policy

Takara Everest Truong, Michael Piseno, Zhaoming Xie, C. Karen Liu

SIGGRAPH AsiaAcademic49 citesMotion Synthesis

Diffusion policy for physics-based animation trained with RL corrective actions, demonstrated on perturbation recovery and text-to-motion tracking.

Abstract

Generating diverse and realistic human motion that can physically interact with an environment remains a challenging research area in character animation. Meanwhile, diffusion-based methods, as proposed by the robotics community, have demonstrated the ability to capture highly diverse and multi-modal skills. However, naively training a diffusion policy often results in unstable motions for high-frequency, under-actuated control tasks like bipedal locomotion due to rapidly accumulating compounding errors, pushing the agent away from optimal training trajectories. The key idea lies in using RL policies not just for providing optimal trajectories but for providing corrective actions in sub-optimal states which gives the policy a chance to correct for errors caused by environmental stimulus, model errors, or numerical errors in simulation. Our method, Physics-Based Character Animation via Diffusion Policy (PDP), combines reinforcement learning (RL) and behavior cloning (BC) to create a robust diffusion policy for physics-based character animation. We demonstrate PDP on perturbation recovery, universal motion tracking, and physics-based text-to-motion synthesis.

How to read this

Category
Method: a diffusion policy for physics-based animation
Contributions
  • PDP, a diffusion policy for physics-based character animation combining reinforcement learning and behavior cloning
  • Uses RL policies to provide corrective actions in sub-optimal states, countering compounding errors in under-actuated control
  • Demonstrated on perturbation recovery, universal motion tracking, and physics-based text-to-motion synthesis
Context
Brings robotics-style diffusion policies into physics-based character control, relating to motion diffusion models such as Tevet et al.'s Human Motion Diffusion Model.Builds on: Human Motion Diffusion Model
Correctness
Rests on the idea that RL-provided corrective actions stabilize an otherwise unstable diffusion policy for high-frequency control; results are shown on the three named tasks, so a reader should note stability still depends on the RL correction design.
Clarity
Accessible if you know diffusion policies; a first pass conveys the RL-plus-BC insight, a second pass for the corrective-action training detail.
How to read it
Read why naive diffusion policies are unstable and how corrective actions fix it; a second pass pays off for the RL/BC combination and the per-task setups.

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 →