Back to search

Cornell University

Imitation Learning for Stylized Physics-Based Character Control

Abstract

dc:description.abstract

In Computer Graphics, a heavily researched topic is the physical simulation of characters that can exhibit fluid, life-like motions. In recent years, imitation and reinforcement learning techniques have become popular approaches for training such controllers due to their flexibility, generality, and adaptability. One such example is DeepMimic, a data-driven framework that utilizes motion clips along with modern reinforcement learning methods to train control policies for simulated characters that can produce a wide variety of natural notions. Adversarial Motion Priors is an extension upon this framework in which adversarial imitation learning is utilized to enable characters to imitate various motions from a large unstructured dataset of reference motions without the need for explicit synchronization. In this thesis, we adapt the Adversarial Motion Priors framework to be compatible with OpenAI Gym environments. In doing so, a wide variety of RL algorithms can be tested on the framework. Finally, we demonstrate the use of this environment by evaluating Model-based Imitation Learning which is a purely offline imitation learning algorithm that tackles the covariate shift issue common in behavior cloning, a classic offline imitation learning algorithm.

Degree

thesis:*
Name thesis:degree_name
M.S., Computer Science
Level thesis:degree_level
Master of Science
Discipline thesis:degree_discipline
Computer Science
Grantor
Cornell University
Year dc:date.issued
2022

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Tsao, Albert
Chair dc:contributor.chair
  • Sun, Wen
Committee member dc:contributor.committeemember
  • Kleinberg, Robert

Subjects

dc:subject × 3

Rights

dc:rights
Statement dc:rights
  • Attribution 4.0 International
Language dc:language.iso
en

Identifiers

dc:identifier.*
Dc Identifier Bibid
15644108
Dc Identifier Other
ProQuest Submission ID: 11579
ProQuest Publication ID: 29395501

Chain of custody

source
Harvested from
Cornell University
Base URL
ecommons.cornell.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Tsao, Albert. Imitation Learning for Stylized Physics-Based Character Control. Master of Science thesis, Cornell University, 2022. https://hdl.handle.net/1813/113052