Learned Motion Matching

This asset generates realistic and smooth transitions for basic humanoid locomotion.

Links:

Demo | Documentation | Website | Paper

Please find the Demo in the "Try it out" section of the website.

Key Features:

• 🏃 Natural, Fluid Movement: Achieve smooth, weight-shifted motion with no foot sliding or abrupt animation changes like in State Machine animations. The customizable kNN classifier balances between fast, responsive transitions and realistic movements.

• 📍 Waypoint Navigation: With an advanced Distance Matching system, powered by a Greedy Algorithm, your characters can follow custom-defined paths with precision. Loop waypoints effortlessly and adjust the speed to control how your character interacts with the environment.

• 🦾 Upper Body Customization: Bring hand interactions to life! Using Inverse Kinematics (IK), your characters can dynamically hold, grasp, or manipulate objects with realistic hand poses. Customize hand weights, fine-tune motions, and create complex finger movements with ease.

• ⚙️ Effortless Setup: Designed for all humanoid characters, with a one-click setup process and comprehensive documentation to get you up and running in no time.

Dependencies:

Learned Motion Matching requires the Animation Rigging and Sentis package from the package manager.

Limitations:

• Locomotion Only: The package is focused solely on basic human locomotion; advanced moves like jumping or vaulting are not included.
• No Specialized Motions: Custom movements, such as limping or skipping, require external data and it is not included in this package.
• Requires Sentis Version 1.3 which is an older version of Sentis.

Research:

This asset implements the SSIGRAPH technical paper <Taming Diffusion Probabilistic Models for Character Control>.

Citation:

Chen, R., Shi, M., Huang, S., Tan, P., Komura, T., & Chen, X. (2024, April 23). Taming diffusion probabilistic models for character control. arXiv.org. https://arxiv.org/abs/2404.15121