Efficient Hyperparameter Optimization for Physics-based Character Animation

被引:3
作者
Yang, Zeshi [1 ]
Yin, Zhiqi [1 ]
机构
[1] Simon Fraser Univ, Burnaby, BC, Canada
关键词
Physics-based Character Animation; Bayesian Optimization; Reinforcement Learning; Curriculum Learning; DESIGN;
D O I
10.1145/3451254
中图分类号
TP31 [计算机软件];
学科分类号
081202 ; 0835 ;
摘要
Physics-based character animation has seen significant advances in recent years with the adoption of Deep Reinforcement Learning (DRL). However, DRL-based learning methods are usually computationally expensive and their performance crucially depends on the choice of hyperparameters. Tuning hyperparameters for these methods often requires repetitive training of control policies, which is even more computationally prohibitive. In this work, we propose a novel Curriculum-based Multi-Fidelity Bayesian Optimization framework (CMFBO) for efficient hyperparameter optimization of DRL-based character control systems. Using curriculum-based task difficulty as fidelity criterion, our method improves searching efficiency by gradually pruning search space through evaluation on easier motor skill tasks. We evaluate our method on two physics-based character control tasks: character morphology optimization and hyperparameter tuning of DeepMimic. Our algorithm significantly outperforms state-of-the-art hyperparameter optimization methods applicable for physics-based character animation. In particular, we show that hyperparameters optimized through our algorithm result in at least 5x efficiency gain comparing to author-released settings in DeepMimic.
引用
收藏
页数:19
相关论文
共 89 条
  • [1] Diverse Motions and Character Shapes for Simulated Skills
    Agrawal, Shailen
    Shen, Shuo
    van de Panne, Michiel
    [J]. IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS, 2014, 20 (10) : 1345 - 1355
  • [2] [Anonymous], 2014, ARXIV14063896
  • [3] [Anonymous], 2017, ABS170702286 CORR
  • [4] Bai Y, 2019, PYBULLET PYTHON MODU
  • [5] Bengio Y., 2009, P 26 ANN INT C MACH, DOI DOI 10.1145/1553374.15533802,5
  • [6] DReCon: Data-Driven Responsive Control of Physics-Based Characters
    Bergamin, Kevin
    Clavet, Simon
    Holden, Daniel
    Forbes, James Richard
    [J]. ACM TRANSACTIONS ON GRAPHICS, 2019, 38 (06):
  • [7] Terrain-Adaptive Locomotion Skills Using Deep Reinforcement Learning
    Bin Peng, Xue
    Berseth, Glen
    van de Panne, Michiel
    [J]. ACM TRANSACTIONS ON GRAPHICS, 2016, 35 (04):
  • [8] Morphological change in machines accelerates the evolution of robust behavior
    Bongard, Josh
    [J]. PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA, 2011, 108 (04) : 1234 - 1239
  • [9] Brochu Eric, 2007, SIGGRAPH POSTERS, V105
  • [10] Brochu Eric, 2010, Proceedings of the 2010 ACM SIGGRAPH/Eurographics Symposium on Computer Animation, P103