Procedural Animation / Behavior

Animation & Motion

Open Source

• DE

  DeVI Discovered by AI

  2026

  A framework that leverages 2D synthetic videos from generative models to train physics-based characters to perform complex, dexterous human-object interactions in 3D physical simulations.

  interaction physics rl video

  arXiv

• GL

  GPC: Large-Scale Generative Pretraining for Transferable Motor Control Discovered by AI

  2026

  Generative Pretrained Controllers (GPC) leverage tokenization and next-token modeling to create general-purpose, reusable generative controllers from large-scale motion datasets using reinforcement learning.

  physics procedural rl transformer

  Project · arXiv · Paper

• MA

  MaskAdapt Discovered by AI

  2026

  A two-stage residual learning framework for flexible motion adaptation in physics-based humanoid control, training a mask-invariant base policy to handle missing observations and adapt to diverse environments.

  adaptation humanoid physics rl

  GitHub · arXiv

• NC

  NaP-Control Discovered by AI

  2026

  A reinforcement learning-driven control method that manipulates the latent noise of a task-agnostic diffusion policy prior, enabling versatile, fast, and robust whole-body character control in physics-based animation.

  character-control diffusion-prior physics-simulation reinforcement-learning

  arXiv

• SW

  SWIM Discovered by AI

  2026

  A reinforcement learning-based method for synthesizing physically-based swimming motions, enabling virtual characters to generate physically valid, controllable, and natural-looking swimming behaviors in complex fluid environments.

  fluid physics rl swimming

  arXiv

• TR

  TextOp: Real-time Interactive Text-Driven Humanoid Robot Motion Generation and Control Discovered by AI

  2026

  A real-time text-driven humanoid motion generation and control framework that supports streaming language commands and on-the-fly instruction modification. It combines a high-level motion diffusion model for continuous short-horizon kinematic trajectory generation with a low-level reinforcement learning-based tracking policy.

  physics-based real-time reinforcement-learning text-to-motion

  Project · arXiv

• AM

  AMOR Discovered by AI

  2025

  A multi-objective reinforcement learning (MORL) framework for training physics-based character controllers conditioned on reward weights, eliminating the need for tedious manual reward tuning.

  character-control physics-simulation reinforcement-learning

  Project · arXiv · Paper

• ML

  Mimic2DM: Learning to Control Physically-simulated 3D Characters via Generating and Mimicking 2D Motions Discovered by AI

  2025

  A motion imitation framework that learns physics-based 3D character control policies directly from widely available 2D keypoint trajectories extracted from videos. It features a view-agnostic tracking policy that can be zero-shot adapted to multi-view tracking and integrated with 2D motion generators for diverse motion synthesis.

  2d-to-3d motion-imitation physics-based reinforcement-learning

  Project · GitHub · arXiv

• MI

  MimicKit Discovered by AI

  2025

  A lightweight suite of motion imitation methods for training motion controllers, supporting algorithms like DeepMimic, AMP, ASE, and SMP across multiple simulation engines.

  character-control motion-imitation physics reinforcement-learning

  GitHub

• PP

  PARC: Physics-based Augmentation with Reinforcement Learning for Character Controllers Discovered by AI

  2025

  A framework that uses reinforcement learning to augment character controllers with agile, physics-based terrain traversal behaviors (such as parkour) without requiring extensive motion capture data.

  parkour physics-based reinforcement-learning

  GitHub · arXiv

• CA

  CAMDM Discovered by AI

  2024

  A real-time character control framework that utilizes a transformer-based Conditional Autoregressive Motion Diffusion Model to generate high-quality and diverse animations in Unity.

  character-control diffusion-model unity

  GitHub · arXiv · Paper

• CC

  CLoSD: Closing the Loop between Simulation and Diffusion for Multi-Task Character Control Discovered by AI

  2024

  A text-driven reinforcement learning framework that guides physics-based character controllers using motion diffusion models, combining the stylistic variety of diffusion with the physical plausibility and environmental interaction of simulation.

  diffusion-policy physics-based reinforcement-learning

  GitHub · arXiv

• FR

  FürElise Discovered by AI

  2024

  A system combining diffusion models and reinforcement learning to synthesize physically plausible, highly dexterous hand motions for piano playing from musical scores.

  hand-animation physics-simulation reinforcement-learning

  Project · arXiv · Paper

• GE

  Genesis Discovered by AI

  2024

  An open-source, high-performance physics simulation platform designed for general-purpose robotics, computer graphics, and physical AI. It features a unified multi-physics engine, a photorealistic renderer, and a cross-platform compiler, achieving simulation speeds significantly faster than existing GPU-accelerated stacks.

  physics rl robotics simulation

  GitHub · Website

• LU

  Learning Uniformly Distributed Embedding Clusters of Stylistic Skills for Physically Simulated Characters Discovered by AI

  2024

  Proposes a skill-conditioned controller that leverages Neural Collapse and an Embedding Expansion technique to learn uniformly distributed stylistic embedding clusters on a hypersphere for physics-based character control.

  physics-simulation reinforcement-learning style-control

  arXiv · Website

• PR

  ProtoMotions Discovered by AI

  2024

  A GPU-accelerated simulation and learning framework developed by NVIDIA for training physically simulated digital humans and humanoid robots using reinforcement learning.

  humanoid physics-simulation reinforcement-learning

  GitHub

• SU

  SuperPADL Discovered by AI

  2024

  An advanced framework that scales language-directed physics-based character control to hundreds of motions using progressive supervised distillation, enabling robust and responsive text-to-behavior synthesis.

  language-directed physics-simulation supervised-distillation text-to-behavior

  Project · arXiv

• AM

  AMDM Discovered by AI

  2023

  An auto-regressive motion diffusion model framework in PyTorch that supports real-time character control, motion inpainting, and reinforcement learning-based interactive control.

  character-control diffusion-model reinforcement-learning

  GitHub · arXiv · Paper

• BA

  Barkour Discovered by AI

  2023

  A benchmark and control framework developed by Google DeepMind to quantify and achieve animal-level agility in quadruped robots. It utilizes deep reinforcement learning to train policies capable of navigating complex obstacle courses with high-speed leaps, climbs, and turns.

  agility benchmark quadruped rl

  GitHub · arXiv

• CA

  CALM Discovered by AI

  2023

  An approach for generating diverse and directable behaviors for user-controlled interactive virtual characters by learning a conditional adversarial latent space of human motion.

  character-control physics reinforcement-learning

  Project · GitHub · arXiv

• DI

  DiffMimic Discovered by AI

  2023

  A motion mimicking framework that leverages differentiable physics simulators to significantly accelerate the training of physically simulated characters.

  character-control differentiable-physics physics

  GitHub · arXiv · Website

• IS

  InterPhys: Synthesizing Physical Character-Scene Interactions Discovered by AI

  2023

  InterPhys uses adversarial imitation learning and reinforcement learning to train physically simulated characters capable of realistic, complex physical interactions with 3D scene objects.

  character-control physics-simulation reinforcement-learning scene-interaction

  arXiv · Paper

• MU

  MoConVQ: Unified Physics-Based Motion Control via Scalable Discrete Representations Discovered by AI

  2023

  A unified framework for physics-based motion control that leverages vector quantized variational autoencoders (VQ-VAE) and reinforcement learning to learn robust motion embeddings from large, unstructured datasets.

  physics-based reinforcement-learning vq-vae

  GitHub · arXiv

• PA

  PADL Discovered by AI

  2023

  A language-directed physics-based character control framework that leverages natural language processing to allow users to direct the behaviors of physically simulated characters.

  character-control language-directed physics-simulation reinforcement-learning

  Project · GitHub · arXiv

• PP

  PhysDiff: Physics-Guided Human Motion Diffusion Model Discovered by AI

  2022

  A physics-guided human motion diffusion model that incorporates physical constraints into the diffusion process via a physics-based motion projection module, eliminating artifacts like floating and foot sliding.

  diffusion physics procedural

  Project · arXiv

• UP

  Unity Procedural Animation Discovered by AI

  2022

  An open-source Unity project demonstrating procedural character locomotion and climbing animations using Inverse Kinematics (IK) in real-time.

  inverse-kinematics procedural-animation unity

  GitHub

• WT

  Walk These Ways Discovered by AI

  2022

  A reinforcement learning framework for quadruped locomotion that learns a single policy encoding a structured family of diverse locomotion strategies. It provides sim-to-real training and deployment tools, allowing rapid behavior tuning and adaptation in real-world environments.

  locomotion quadruped rl sim-to-real

  GitHub · arXiv

• UU

  UniCon: Universal Neural Controller For Physics-based Character Motion Discovered by AI

  2020

  A physics-based universal neural controller that learns to master thousands of motions with different styles by training on large-scale motion datasets using reinforcement learning.

  physics procedural rl

  Project · arXiv

• AI

  AI4Animation Discovered by AI

  2018

  A comprehensive framework and repository for deep learning-based character animation and control in Unity, supporting biped/quadruped locomotion and neural animation layering.

  character-control locomotion neural-animation unity

  GitHub

• DE

  DeepMimic Discovered by AI

  2018

  A foundational framework that uses deep reinforcement learning to enable physically simulated characters to robustly imitate a wide variety of highly dynamic example motion clips.

  character-control physics reinforcement-learning

  Project · GitHub · arXiv

• UM

  Unity ML-Agents Discovered by AI

  2017

  An open-source toolkit that enables games and simulations to serve as environments for training intelligent agents using deep reinforcement learning and imitation learning.

  agent-training imitation-learning reinforcement-learning unity

  GitHub

Proprietary / Industry Tools

• MP

  Massive Prime Discovered by AI

  2024

  Massive Prime is an industry-standard crowd simulation and procedural character behavior tool that uses autonomous AI agents with neural decision-making and dynamic ragdoll physics for film, TV, and games.

  ai-agents crowd-simulation procedural-behavior ragdoll-physics

  Website

• PU

  PuppetMaster Discovered by AI

  2015

  An advanced physics-based ragdoll and procedural animation tool for Unity that allows developers to animate ragdolls, create biped ragdolls, and simulate realistic physical interactions.

  behavior procedural

  Website

• NE

  NaturalMotion Euphoria Discovered by AI

  2006

  A legendary game animation middleware based on Dynamic Motion Synthesis (DMS) that simulates 3D character biomechanics and physics in real-time.

  middleware physics procedural-animation

  Website