모션 합성 데이터를 활용한 강화학습 기반 객체 반응형 애니메이션 생성
Object Reactive Animation Generation Using Reinforcement Learning Based on Motion Synthesis Data
1
조회
0
다운로드
1
인용
0
참고문헌
키워드
초록
애니메이션은 주어진 키프레임(key frame)에 맞춰 움직이기 때문에, 외부 객체(object)와 상호작용하기 위해서는 객체의 위치와 방향에 맞춰서 애니메이션을 변환해야 한다. 이 논문에서는 기존 애니메이션을 합성한 데이터로 애니메이션을 생성하고, 생성한 애니메이션으로 객체와의 상호작용을 학습하여 환경에 최적화된 애니메이션을 생성하는 방법을 제안한다. 애니메이션을 학습하는 에이전트(agent)는 개체의 상태를 관찰하고 주어진 애니메이션들의 움직임을 합성하여 다양한 행동을 만들어내고 목적을 달성한 정도에 따라 보상을 받는다. 우리는 에이전트가 행동하는 과정에서 모션 합성 비율에 대한 가중치를 계산하는 함수를 설계하고, 생성되는 애니메이션이 사람이 취할 수 있는 동작에 가깝도록 회전 각도 clamping 보간 함수와 보상 시스템을 설계하여 반영하였다. 모션 합성 데이터를 활용한 강화학습 기반 애니메이션 생성 기법은 객체의 변화에 반응하는 애니메이션을 기존 강화학습 기반 기법보다 적은 학습량으로 생성할 수 있음을 확인했다.
Abstract
In animation, it is necessary to modify the animation to align with the position and rotation of external objects, as animations operate according to given keyframes. This paper proposes a method to generate animations using synthesis data from existing animations and to learn environment-optimized animations through interactions with objects. The agent learning the animation observes the state of the object and synthesis motions from given animations to perform various actions, receiving rewards based on the degree of goal achievement. We design a function to calculate the weight of motion synthesis during the agent's actions and incorporate a rotation angle clamping interpolation function and a reward system to ensure that the generated animations closely resemble human movements. The reinforcement learning-based animation generation technique using motion synthesis data is found to generate animations interacting with changing objects with less training than traditional reinforcement learning-based methods.
참고문헌 (27)
- Sutton, Richard S., "Reinforcement learning: An introduction" (2018), MIT press
- Bellman, Richard, "A Markovian decision process", Journal of mathematics and mechanics, pp.679~684 (1957)
- Peng, Xue Bin, "Deepmimic : Exampleguided deep reinforcement learning of physicsbased character skills", ACM Transactions On Graphics (TOG), 37(4), pp.1~14 (2018)
- Wang, Jack M., "Optimizing locomotion controllers using biologically-based actuators and objectives", ACM Transactions on Graphics(TOG), 31(4), pp.1~11 (2012)
- Lee, Seunghwan, "Dexterous manipulation and control with volumetric muscles", ACM Transactions on Graphics(TOG), 37(4), pp.1~13 (2018)
- Lee, Yoonsang, "Locomotion control for many-muscle humanoids", ACM Transactions on Graphics(TOG), 33(6), pp.1~11 (2014)
- Delicado, Luis, "An automatic tool to facilitate authoring animation blending in game engines", Proceedings of the 12th ACM SIGGRAPH Conference on Motion, Interaction and Games (2019)
- Aini, Afifah Nur, "Comparison of Inverse Kinematics and Forward Kinematics Methods on Walk Cycle Animation Characters", Telematika, 14(1), pp.55~63 (2021)
- Lewis, John P., "Direct manipulation blendshapes", IEEE Computer Graphics and Applications, 30(4), pp.42~50 (2010)
- Barr, Alan H., "Smooth interpolation of orientations with angular velocity constraints using quaternions", ACM SIGGRAPH Computer Graphics, 26(2), pp.313~320 (1992)
- Jacobson, Alec, "Fast automatic skinning transformations", ACM Transactions on Graphics(TOG)31, 31(4), pp.1~10 (2012)
- Kavan, Ladislav, "Geometric skinning with approximate dual quaternion blending", ACM Transactions on Graphics (TOG), 27(4), pp.1~23 (2008)
- Kavan, Ladislav, "Skinning with dual quaternions", Proceedings of the 2007 symposium on Interactive 3D graphics and games (2007)
- Arikan, Okan, "Motion synthesis from annotations", ACM SIGGRAPH 2003, pp.402~408 (2003)
- Bruderlin, Armin, "Motion signal processing", Proceedings of the 22nd annual conference on Computer graphics and interactive techniques (1995)
- Park, Sang Il, "On-line locomotion generation based on motion blending", Proceedings of the 2002 ACM SIGGRAPH, 2002, ~
- Rose, Charles, "Verbs and adverbs: Multidimensional motion interpolation", IEEE Computer Graphics and Applications, 18(5), pp.32~40 (1998)
- Yuan, Ye, "Simpoe: Simulated character control for 3d human pose estimation", Proceedings of the IEEE, 2021, ~
- Lee, Kyungho, "Interactive character animation by learning multi-objective control", ACM Transactions on Graphics (TOG), 37(6), pp.1~10 (2018)
- Liu, Libin, "Learning basketball dribbling skills using trajectory optimization and deep reinforcement learning", ACM Transactions on Graphics (TOG), 37(4), pp.1~14 (2018)
- Duan, Yan, "Benchmarking deep reinforcement learning for continuous control", International conference on machine learning (2016), PMLR
- Peng, Xue Bin, "Terrain-adaptive locomotion skills using deep reinforcement learning", ACM Transactions on Graphics (TOG), 35(4), pp.1~12 (2016)
- Peng, Xue Bin, "Amp : Adversarial motion priors for stylized physics-based character control", ACM Transactions on Graphics (ToG), 40(4), pp.1~20 (2021)
- Yuan, Ye, "Physdiff: Physics-guided human motion diffusion model", Proceedings of the IEEE, 2023, ~
- "Unity-Technologies/ML-agents"[URL]