Deep Learning architectures for generating human motion

Abstract

Human motion generation is a critical area of research with applications spanning animation, virtual reality, robotics, and healthcare. It involves creating realistic and dynamic representations of human movements through computational models. Traditional methods often struggle to capture the complexity and variability of natural human motion. Recent advancements in deep learning offer promising solutions, leveraging sophisticated neural network architectures that excel in learning temporal sequences and generating high-fidelity motion data. Despite these advancements, the field continues to grapple with issues such as data scarcity, the need for extensive computational resources, and ensuring the generalization of models across different types of motion. This thesis delves into these challenges, exploring the latest deep-learning techniques that aim to enhance the realism and diversity of human motion generation.