Modularized Predictive Coding-Based Online Motion Synthesis Combining Environmental Constraints and Motion-Capture Data

Abstract

Motion synthesis benefits from the use of motion capture data and a dynamic model because the motion data can provide a reference to naturalness, and the dynamic model can support environmental constraints such as footskate prevention or perturbation response. However, a combination of a dynamic model and captured motion usually demands professional insights, experience, and additional efforts such as preprocessing or off-line optimization. To address this issue, we propose a modularized predictive coding-based motion synthesis framework that synthesizes natural motion while maintaining the constraints. Modularized predictive coding provides intuitive online mediation of multiple information sources, which can then be applied to motion synthesis. To validate the proposed framework, we applied different types of motion data and character models to synthesize human walking, kickboxing, and backflipping motions, a dog walking motion, and a hand object-grasping motion.

Keywords

• Combination of linear models
• hybrid-based character animation
• neuroscience-inspired
• online motion synthesis