基於音樂分析之人形機器人的平滑動作規劃

Abstract

針對人形機器人之娛樂性的應用，本論文提出藉由輸入之音樂種類與其節拍，搭配動作資料庫，使人形機器人有流暢舞蹈動作的軌跡規劃。人形機器人由於擁有似人的外表，在應用上更能貼近人類的行為與生活環境，除了任務性的需求以外，娛樂性也是人形機器人的應用之一，在進行舞蹈動作時，人形機器人要如何有穩定且平滑的軌跡，且能夠搭配音樂的節拍與種類，為本論文主要目的。 在輸入音樂後，會針對種類與節拍作分析，種類會對應到符合的動作資料庫，節拍則會影響軌跡的生成，因此在建立動作資料庫時會需要找出關鍵的通過點，將這些通過點紀錄成一連串的特徵點，藉由特徵點間的時間差異作符合節拍的軌跡規劃，在動作與動作之間也藉由特徵點來作平滑的連接，在產生平滑軌跡後，讓人形機器人達到表演的能力。

Considering the entertainment potential of humanoid robots, we propose a concept of music analysis with a movement database to generate smooth dancing trajectories. Be-cause the robots have an anthropomorphic appearance, they can be similar to human beings in behavior or environment. Other than the task-based needs, entertainment is one of the most popular applications for humanoid robot. The purpose of this thesis is to discover how to achieve stable and smooth trajectories, while considering the beat and the genre of music when dancing. After the music input, music genre classification and beat tracking are executed. Genre classification allows choosing the movements from database accordingly. Furthermore, beat tracking will affect the generation of the trajectory. Therefore, while building the database of movements, it is crucial to find the key points of movements. Afterwards, these key points are saved as feature points to be used for trajectory planning to match the beat. The connections between chains of stored movements are also based on the trajectory planning by feature points. Moreover, the balance of the humanoid robot is significant for a stable performance.