Machine Learning Pose Prediction for Character Animation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current 3D animation software requires extensive manual effort and expertise to pose virtual characters, limiting the ability of non-experts to create high-quality animations, and existing inverse kinematic solvers often produce physically plausible but not naturally looking poses.
Innovation Solution
A machine learning-based pose prediction system that generates animation poses by transforming variable input constraints into a pose embedding, which is then expanded into local rotation and global position data for character joints, allowing for the creation of natural-looking poses without manual intervention or additional training for different character types.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If manual posing by directly manipulating bones is used, then precise control of character skeleton is achieved, but the task becomes painful and time-consuming due to hundreds of bones
Solution Approach 1:
The patent replaces the manual mechanical manipulation of hundreds of individual bones with an automated machine learning system. The ML model automatically computes joint positions and rotations based on effector constraints, substituting the manual mechanical posing process with an automated computational approach that maintains precision while dramatically reducing time investment.
Solution Approach 2:
The system enables self-service automation where the ML model independently solves the complex pose computation problem without requiring manual intervention. The model takes effector constraints as input and automatically generates the complete character pose, allowing animators to work autonomously without needing to manually adjust each bone.
2Reliability
If inverse kinematic solvers are used to solve for skeleton joint positions, then physically-plausible poses are obtained, but the poses are not necessarily natural-looking and require parameterization
Solution Approach 1:
The patent changes the fundamental parameters of the pose solving approach by using a machine learning model trained on natural motion data rather than traditional IK parameters. The ML model learns the complex relationships between effector positions and natural character poses, producing physically plausible and naturally looking poses simultaneously without requiring manual parameterization of twist or swing constraints.
Solution Approach 2:
The system substitutes the traditional inverse kinematics mathematical solver with a machine learning-based approach. This replacement enables the system to produce natural-looking poses by learning from training data while maintaining physical plausibility, overcoming the limitation of traditional IK solvers that can only guarantee physical correctness but not natural appearance.
3Adaptability or versatility
If traditional rigging with multiple controls is implemented, then expressive character control is achieved, but much work is required to design and parameterize each control
Solution Approach 1:
The patent implements a universal ML-based pose prediction system that can handle multiple types of effector constraints (position, rotation, look-at) through a single unified model. This universal approach eliminates the need to create separate controls for different character parts, as the model generalizes across the entire character skeleton, reducing rigging complexity while maintaining full expressiveness.
Solution Approach 2:
The system extracts the complex rigging and control design tasks from the animation pipeline by using a pre-trained ML model. Instead of requiring animators to design and parameterize individual controls for each bone and joint, the model is extracted as a standalone component that automatically handles pose computation, significantly simplifying the overall system complexity.
4Manufacturing precision
If IK solvers with constraint solving are used, then bone chain positioning is achieved, but the approach is limited to two-bone chains and requires additional parameterization
Solution Approach 1:
The patent creates a universal ML-based pose prediction system that handles complex full-body character poses, going far beyond the two-bone chain limitation of traditional IK solvers. The model can simultaneously compute positions and rotations for all joints in a character skeleton, providing both high precision and broad applicability to entire character rigs rather than isolated bone chains.
Data Source
AI summary
A method of generating or modifying poses in an animation of a character are disclosed. Variable numbers and types of supplied inputs are combined into a single input. The variable numbers and types of supplied inputs correspond to one or more effector constraints for one or more joints of the character. The single input is transformed into a pose embedding. The pose embedding includes a machine-learned representation of the single input. The pose embedding is expanded into a pose representation output. The pose representation output includes local rotation data and global position data for the one or more joints of the character.