Local Motion Phases for Learning Multi-Contact Character Movements, present a novel deep learning framework to synthesize different animations from unstructured motion dataset.
The framework can help learn fast and dynamic character interactions involving multiple contacts between the body and an object or other characters and the environment.
Using a novel concept called local motion phase, the framework was able to produce a variety of motion skills, like ball dribbling.
It could also produce professional maneuvers in basketball like shooting, catching, avoidance, multiple locomotion, different character and object interactions.
Local motion phase concept helps neural networks to learn asynchronous movements.