Pose-aware Neural Inverse Kinematics for Virtual Character Posing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional neural inverse kinematics techniques for posing and animating virtual characters are time-consuming, laborious, and often produce unnatural or unrealistic poses due to global pose changes, which can destroy previous edits and lack efficient preservation of base poses.
Innovation Solution
A pose-aware neural inverse kinematics method using a graph neural network with a cross-layer attention mechanism that selectively preserves aspects of a base pose by generating updated joint states and orientations, allowing for iterative refinement and natural, realistic pose generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If conventional neural IK models generate global poses from sparse control inputs, then the model can compute positions and orientations of un-manipulated joints, but the pose output is influenced by control inputs in unwanted ways causing unnatural pose changes
Solution Approach 1:
The patent segments the pose generation process into local and global components. The neural network processes sparse control inputs to generate local pose adjustments, while a base pose provides the global structure. This segmentation allows the system to maintain automated generation while preventing unwanted global influences on local pose elements.
Solution Approach 2:
The patent extracts the base pose from the neural network output and uses it as a separate reference. By taking out the base pose information and comparing it with the neural network-generated adjustments, the system can preserve the intended base pose structure while incorporating only the necessary local adjustments from control inputs.
2Productivity
If neural IK models generate poses based on sparse control inputs, then pose computation is automated, but previous edits in the base pose are destroyed or changed
Solution Approach 1:
The patent implements feedback by continuously comparing the generated pose with the base pose and adjusting the neural network output accordingly. The system uses the base pose as a reference feedback signal to ensure that generated poses maintain the intended structure while incorporating necessary adjustments from control inputs.
Solution Approach 2:
The patent performs preliminary action by pre-defining the base pose with all intended edits and adjustments before neural network processing. The base pose is prepared in advance with the correct structure, and the neural network only adds necessary local adjustments, ensuring previous edits are preserved rather than destroyed.
3Manufacturing precision
If manual manipulation of control handles is used for posing, then precise control over joints is achieved, but the process is time-consuming and laborious
Solution Approach 1:
The patent replaces the manual mechanical manipulation of control handles with an automated neural network system. The neural network automatically computes joint positions and orientations based on sparse control inputs, eliminating the need for time-consuming manual manipulation while maintaining pose accuracy through the base pose reference.
Data Source
AI summary
One embodiment of the present invention sets forth a technique for generating a pose for a virtual character. The technique includes determining a set of joint representations corresponding to a set of joints in the virtual character based on (i) a base pose for the virtual character and (ii) a set of constraints associated with one or more joints included in the set of joints. The technique also includes generating, via execution of a first neural network, a set of updated joint states for the set of joints based on the set of joint representations. The technique further includes generating, based on the set of updated joint states, an output pose that includes (i) a first set of joint positions for the set of joints and (ii) a first set of joint orientations for the set of joints.