Physics-Based Animation¶
Animation & Motion
Open Source¶
-
BFMTrack: Latent Sequence Optimization for Physics-Based Motion Tracking with Behavioral Foundation Models Discovered by AI
2026
Proposes Latent Sequence Optimization (LSO) to extend Behavioral Foundation Models (BFMs) for precise, time-varying physics-based motion tracking without requiring manual reward engineering.
character-controlfoundation-modelmotion-trackingphysics -
InstantMimic: A High Performance System for Learning Physics-based Skills in Seconds Discovered by AI
2026
A GPU-native reinforcement learning system designed to learn physics-based character skills in seconds. By leveraging high-performance simulation and parallelized training, it enables rapid motion mimicking and makes LLM-agent-driven hyperparameter search practical.
gpu-nativemotion-mimickingphysicsreinforcement-learning -
InterAgent: Physics-based Multi-agent Command Execution via Diffusion on Interaction Graphs Discovered by AI
2026
A text-driven, physics-based dual-humanoid agent control framework that models multi-agent interactions using diffusion on interaction graphs. It enables physically plausible, contact-rich interactions between multiple simulated characters based on textual commands.
diffusionhumanoidmulti-agentphysics -
Neural Assistive Impulses: Synthesizing Exaggerated Motions for Physics-based Characters Discovered by AI
2026
Proposes a method to synthesize stylized and exaggerated motions for physics-based characters by introducing neural assistive impulses that temporarily bypass standard physical constraints.
exaggerated-motionphysics-animationstylized-animation -
PhysMoDPO: Physically-Plausible Humanoid Motion with Preference Optimization Discovered by AI
2026
Introduces a preference optimization framework to align physics-based whole-body controllers with human preferences, correcting physical deviations in diffusion-generated motions while preserving naturalness.
humanoid-motionphysics-animationpreference-optimization -
SCRIPT: Scalable Diffusion Policy with Multi-stage Training for Language-driven Physics-Based Humanoid Control Discovered by AI
2026
A scalable diffusion policy with a multi-stage training framework for language-driven physics-based humanoid control. It features a Joint Action-State-Text Diffusion Transformer (JAST-DiT) that models actions, physical states, and text as dedicated token streams coupled through joint attention.
diffusionhumanoidphysicstransformer -
Tired Actor: Fatigue-Informed Character Control Discovered by AI
2026
A physics-based character animation framework that incorporates fatigue as a physiological prior using the Three-Compartment Controller (3CC) model, enabling characters to adapt their movements naturally under varying energy limits.
biomechanicscharacter-controlfatiguephysics -
BeyondMimic: From Motion Tracking to Versatile Humanoid Control via Guided Diffusion Discovered by AI
2025
A framework that scales to diverse human motions and enables versatile, steerable humanoid control via guided diffusion-based controllers, bridging sim-to-real motion tracking and flexible test-time control.
humanoid-controlmotion-trackingphysics-animation -
Diffuse-CLoC: Guided Diffusion for Physics-based Character Look-ahead Control Discovered by AI
2025
Proposes a guided diffusion framework for physics-based look-ahead control that models the joint distribution of states and actions, enabling steerable, physically realistic motion generation for long-horizon tasks like obstacle avoidance.
character-controldiffusionphysics -
HIL: Hybrid Imitation Learning for Dynamic Athletic Control Discovered by AI
2025
A hybrid imitation learning framework that combines motion tracking for precise skill replication with adversarial imitation learning to train unified controllers for diverse parkour and athletic skills.
imitation-learningparkourphysics -
InterMimic: Towards Universal Whole-Body Control for Physics-Based Human-Object Interactions Discovered by AI
2025
A framework for learning universal whole-body control policies from imperfect motion capture data, enabling physically simulated characters to perform diverse human-object interactions.
character-controlphysicsreinforcement-learning -
KungfuBot Discovered by AI
2025
A physics-based humanoid control framework designed to master highly-dynamic human behaviors such as Kungfu and dancing through multi-step motion processing and adaptive motion tracking.
humanoid-controlmotion-trackingphysics -
MetaMotivo: Zero-Shot Whole-Body Humanoid Control via Behavioral Foundation Models Discovered by AI
2025
A behavioral foundation model pre-trained with unsupervised reinforcement learning to control the movements of a complex virtual humanoid agent. It can solve unseen tasks such as motion tracking, pose reaching, and reward optimization via zero-shot inference without additional fine-tuning.
foundation-modelhumanoidphysicsrl -
MimicKit Discovered by AI
2025
An open-source reinforcement learning framework designed for training physics-based character controllers using motion imitation. It provides modular and highly configurable implementations of various motion-imitation techniques and RL algorithms to support research in computer graphics and robotics.
motion-imitationphysicsreinforcement-learningrobotics -
PhysAnimator Discovered by AI
2025
A physics-guided generative framework that combines 2D deformable body simulations with sketch-guided video diffusion models to animate static anime illustrations. It allows interactive control via customizable energy strokes and rigging points to generate physically plausible and stylized cartoon animations.
animedeformable-bodydiffusionphysics -
2025
Predicts physical material properties from visual input ('physics from pixels') so that generated 3D scenes can be simulated with realistic dynamics.
material-propertiesphysicssimulation -
SAFT: Shape and Appearance of Fabrics from Template
2025
Recovers the shape and appearance of fabrics from a template via differentiable physical simulation on monocular video, yielding physically-grounded cloth assets.
clothdifferentiablesimulation -
SkillMimic-V2: Learning Robust and Generalizable Interaction Skills from Sparse and Noisy Demonstrations Discovered by AI
2025
A reinforcement learning framework that enables physically simulated characters to learn robust and generalizable interaction skills from sparse, noisy, and disconnected demonstrations.
character-controlinteraction-skillsphysics-animationreinforcement-learning -
SMP: Reusable Score-Matching Motion Priors for Physics-Based Character Control Discovered by AI
2025
A framework that leverages pre-trained motion diffusion models and score distillation sampling (SDS) to create reusable, task-agnostic motion priors for physics-based character control.
diffusionmotion-priorphysics -
UniPhys: Unified Planner and Controller with Diffusion for Flexible Physics-Based Character Control Discovered by AI
2025
A unified diffusion-based behavior cloning framework that integrates motion planning and low-level physics control into a single model, enabling zero-shot generalization to diverse tasks.
diffusionmotion-planningphysics -
Versatile Physics-based Character Control with Hybrid Latent Representation Discovered by AI
2025
A physics-based character control framework that integrates continuous and discrete latent representations to build a versatile motion prior shared across multiple challenging tasks.
isaac-gymlatent-representationphysics -
VideoMimic Discovered by AI
2025
A real-to-sim-to-real pipeline that reconstructs human motion and environment context from monocular videos to train physics-based, context-aware humanoid control policies.
humanoidimitation-learningphysics -
Masked Humanoid Controller (MHC) Discovered by AI
2024
A learned whole-body controller that exposes a simple yet expressive interface for commanding diverse humanoid behaviors by specifying masked target trajectories over selected subsets of state variables.
controlhumanoidphysicswhole-body -
MaskedMimic: Unified Physics-Based Character Control Through Masked Motion Inpainting Discovered by AI
2024
Formulates physics-based character control as a general motion inpainting problem, allowing characters to be controlled via sparse keyframes, text instructions, or scene constraints while maintaining physical plausibility.
character-controlmotion-inpaintingphysics-animation -
PDP: Physics-Based Character Animation via Diffusion Policy Discovered by AI
2024
Combines reinforcement learning and behavior cloning to train a robust diffusion policy for physics-based character animation, enabling stable bipedal locomotion, perturbation recovery, and text-to-motion synthesis.
diffusion-policylocomotionphysics-animation -
2024
Learns the physics of dressed 3D avatars from multi-view video, recovering body, garments, and cloth dynamics for physically realistic, animatable avatars.
avatarclothphysics -
Strategy and Skill Learning for Physics-based Table Tennis Animation Discovered by AI
2024
A physics-based table tennis animation framework that uses deep reinforcement learning to enable virtual characters to learn complex table tennis strategies and motor skills, allowing them to play dynamically with other agents or human players in VR.
character-controlphysicsreinforcement-learningtable-tennis -
AdaptNet: Policy Adaptation for Physics-Based Character Control Discovered by AI
2023
A reinforcement learning framework that modifies the latent space of existing policies to allow physically simulated characters to quickly adapt and learn new stylistic skills.
character-controlphysicsreinforcement-learning -
C-ASE: Learning Conditional Adversarial Skill Embeddings for Physics-based Characters Discovered by AI
2023
C-ASE is an efficient framework that learns conditional Adversarial Skill Embeddings for physics-based characters. It divides heterogeneous skill motions into distinct subsets containing homogeneous samples to train a low-level conditional model, enabling physically simulated characters to learn a diverse repertoire of skills with direct controllability.
adversarial-learningcharacter-controlphysics -
InsActor: Instruction-driven Physics-based Characters Discovered by AI
2023
InsActor is a generative framework that leverages diffusion-based human motion models to produce instruction-driven animations of physics-based characters. It employs a hierarchical design with a high-level planner and a low-level controller to generate physically simulated motions from language and waypoint conditions.
character-controldiffusion-policyphysics -
Neural Categorical Priors for Physics-Based Character Control Discovered by AI
2023
A learning framework that uses reinforcement learning and discrete information bottlenecks to learn reusable motion priors, significantly improving motion quality and diversity for physics-based characters.
motion-priorphysicsreinforcement-learning -
UniHSI Discovered by AI
2023
UniHSI is a unified human-scene interaction framework that uses language commands and contact-chain prompts to synthesize physics-grounded whole-body character movements in complex 3D environments.
character-controlhuman-scene-interactionphysicsreinforcement-learning -
ASE: Large-Scale Reusable Adversarial Skill Embeddings for Physically Simulated Characters Discovered by AI
2022
ASE (Adversarial Skill Embeddings) is a large-scale data-driven framework for learning versatile and reusable skill embeddings for physically simulated characters. By combining adversarial imitation learning with unsupervised reinforcement learning, it enables characters to acquire a diverse repertoire of skills that can be easily directed to perform downstream tasks.
character-controlphysicsreinforcement-learning -
ControlVAE: Model-Based Learning of Generative Controllers for Physics-Based Characters Discovered by AI
2022
ControlVAE is a model-based framework for learning generative motion control policies based on variational autoencoders (VAE). It learns a rich and flexible latent representation of skills and a skill-conditioned generative control policy from unorganized motion sequences, enabling realistic physics-based character control.
character-controlphysicsvae -
AMP Discovered by AI
2021
An adversarial reinforcement learning framework for training stylized physics-based character controllers without requiring hand-crafted imitation rewards. It utilizes a discriminator trained on reference motion data to guide the character toward natural, high-fidelity movements.
adversarial-learningcharacter-controlmotion-priorphysics
Proprietary / Industry Tools¶
-
Ragdoll Dynamics Discovered by AI
2021
A real-time physics solver for Autodesk Maya and Blender that allows animators to generate overlapping animation and full-body dynamics directly on animation controls.
blender-addonmaya-pluginphysics-animationreal-time-physics -
DeepMotion Discovered by AI
2018
An AI-driven motion intelligence platform that provides physics-based motion synthesis and video-to-3D animation tools. It enables digital characters to interact responsively with physical environments, reducing foot-sliding and ensuring physically plausible movements.
animationmocapphysicssimulation -
A physics-aware keyframe animation tool with AI auto-posing and interpolation, designed for believable character action animation.
commercialkeyframephysics
