확률적 모델예측제어를 이용한 물리기반 제어기 지도 학습 프레임워크
A Supervised Learning Framework for Physics-based Controllers Using Stochastic Model Predictive Control
0
조회
0
다운로드
1
인용
0
참고문헌
키워드
초록
본 논문에서는 확률적 모델예측제어(model predictive control) 기법을 이용하여 예제 동작 데이터가 주어지면 물리 기반 시뮬레이션 환경에서 그 동작을 모방할 수 있는 캐릭터 동작 제어기를 빠르게 학습할 수 있는 간편한 지도 학습(supervised learning) 프레임워크를 제안한다. 제안된 프레임워크는 크게 학습 데이터 생성과 오프라인 학습의 두 컴포넌트로 구성된다. 첫번째 컴포넌트는 예제 동작 데이터가 주어지면 확률적 모델예측제어를 통해 그 동작 데이터를 추적하기 위한 최적 제어기를 캐릭터의 현재 상태로부터 시작하여 가까운 미래 상태까지의 시간 윈도우에 대해 주기적으로 업데이트하면서 그 최적 제어기를 통해 캐릭터의 동작을 확률적으로 제어한다. 이러한 주기적인 최적 제어기의 업데이트와 확률적 제어는 주어진 예제 동작 데이터를 모방하는 동안 캐릭터가 가질 수 있는 다양한 상태들을 효과적으로 탐색하게 하여 지도 학습에 유용한 학습 데이터를 수집할 수 있게 해준다. 이렇게 학습 데이터가 수집되면, 오프라인 학습 컴포넌트에서는 그 수집된 데이터를 정규화 시켜서 데이터에 내제된 크기와 단위의 차이를 조정하고 지도 학습을 통해 제어기를 위한 간단한 구조의 인공 신경망을 학습시킨다. 걷기 동작과 달리기 동작에 대한 실험은 본 논문에서 제안한 학습 프레임워크가 물리 기반 캐릭터 동작 제어기를 빠르고 효과적으로 생성할 수 있음을 보여준다.
Abstract
In this paper, we present a simple and fast supervised learning framework based on model predictive control so as to learn motion controllers for a physic-based character to track given example motions. The proposed framework is composed of two components: training data generation and offline learning. Given an example motion, the former component stochastically controls the character motion with an optimal controller while repeatedly updating the controller for tracking the example motion through model predictive control over a time window from the current state of the character to a near future state. The repeated update of the optimal controller and the stochastic control make it possible to effectively explore various states that the character may have while mimicking the example motion and collect useful training data for supervised learning. Once all the training data is generated, the latter component normalizes the data to remove the disparity for magnitude and units inherent in the data and trains an artificial neural network with a simple architecture for a controller. The experimental results for walking and running motions demonstrate how effectively and fast the proposed framework produces physics-based motion controllers.
참고문헌 (25)
- Peng, Xue Bin, "Deeploco: Dynamic locomotion skills using hierarchical deep reinforcement learning", ACM Transactions on Graphics (TOG), 36(4), pp.1~13 (2017)
- Peng, Xue Bin, "Deepmimic: Example-guided deep reinforcement learning of physics-based character skills", ACM Transactions on Graphics (TOG), 37(4), pp.1~14 (2018)
- Lee, Seunghwan, "Scalable muscle-actuated human simulation and control", ACM Transactions on Graphics (TOG), 38(4), pp.1~13 (2019)
- Won, Jungdam, "A scalable approach to control diverse behaviors for physically simulated characters", ACM Transactions on Graphics (TOG), 39(4), pp.33~31 (2020)
- Merel, Josh, "Catch & Carry: reusable neural controllers for vision-guided whole-body tasks", ACM Transactions on Graphics (TOG), 39(4), pp.39~31 (2020)
- Han, Daseong, "Data‐guided model predictive control based on smoothed contact dynamics", Computer Graphics Forum, 35(2), pp.533~543 (2016)
- da Silva, Marco, "Interactive simulation of stylized human locomotion", ACM SIGGRAPH 2008 papers, pp.1~10 (2008)
- Mordatch, Igor, "Robust physics-based locomotion using low-dimensional planning", ACM SIGGRAPH 2010 papers, pp.1~8 (2010)
- Tassa, Yuval, "Synthesis and stabilization of complex behaviors through online trajectory optimization", 2012 IEEE, 2012, 4906~4913
- Hämäläinen, Perttu, "Online motion synthesis using sequential monte carlo", ACM Transactions on Graphics (TOG), 33(4), pp.1~12 (2014)
- Kwon, Taesoo, "Control Systems for Human Running using an Inverted Pendulum Model and a Reference Motion Capture Sequence", Symposium on Computer Animation, pp.129~138 (2010)
- Kwon, Taesoo, "Momentum-mapped inverted pendulum models for controlling dynamic human motions", ACM Transactions on Graphics (TOG), 36(1), pp.1~14 (2017)
- Kwon, Taesoo, "Fast and flexible multilegged locomotion using learned centroidal dynamics", ACM Transactions on Graphics (TOG), 39(4), pp.46~41 (2020)
- Han, Daseong, "On‐line real‐time physics‐based predictive motion control with balance recovery", Computer Graphics Forum, 33(2), pp.245~254 (2014)
- Han, Daseong, "Physics‐based trajectory optimization with automatic time warping", Computer Animation and Virtual Worlds, 29(3-4) (2018)
- 한다성, "분석적으로 미분 가능한 시스템 동역학을 이용한 온라인 동작 합성 기법", 한국컴퓨터그래픽스학회논문지, 25(3), pp.133~142 (2019), (사)한국컴퓨터그래픽스학회[DOI]
- Hong, Seokpyo, "Physics-based full-body soccer motion control for dribbling and shooting", ACM Transactions on Graphics (TOG), 38(4), pp.1~12 (2019)
- Eom, Haegwang, "Model Predictive Control with a Visuomotor System for Physics-based Character Animation", ACM Transactions on Graphics (TOG), 39(1), pp.1~11 (2019)
- Jacobson, David H., "Differential dynamic programming" (1970)
- Levine, Sergey, "Guided policy search", International Conference on Machine Learning, pp.1~9 (2013)
- Zhang, Tianhao, "Learning deep control policies for autonomous aerial vehicles with mpc-guided policy search", 2016 IEEE international conference on robotics and automation (ICRA), pp.528~535 (2016)
- Peng, Xue Bin, "Deepmimic: Example-guided deep reinforcement learning of physics-based character skills", ACM Transactions on Graphics (TOG), 37(4), pp.1~14 (2018)
- Holden, Daniel, "Robust solving of optical motion capture data by denoising", ACM Transactions on Graphics (TOG), 37(4), pp.1~12 (2018)
- Holden, Daniel, "Phase-functioned neural networks for character control", ACM Transactions on Graphics (TOG), 36(4), pp.1~13 (2017)
- Abadi, Martín, "Tensorflow: A system for large-scale machine learning", 12th {USENIX} symposium on operating systems design and implementation, 16, pp.265~283 (2016)