Optimal imitation of human movements by robots using intelligent methods

نوع: Type: thesis

مقطع: Segment: masters

عنوان: Title: Optimal imitation of human movements by robots using intelligent methods

ارائه دهنده: Provider: Zahra Dabiri

اساتید راهنما: Supervisors: Dr.Hassan Khotanlou

اساتید مشاور: Advisory Professors:

اساتید ممتحن یا داور: Examining professors or referees: Dr. Muharram MansooriZadeh - Dr. Hatam Abdoli

زمان و تاریخ ارائه: Time and date of presentation: "September 25, 2021"

مکان ارائه: Place of presentation: "Ms. Mohammadi Engineer Hall, Faculty of Engineering"

چکیده: Abstract: "Robots interact with humans to train and plan for easy communication. There are different types of robots based on their appearance, and one of the types of robot models is the humanoid robot. Anthropomorphic robots can imitation human behavior according to the learning model. This imitation can be effective in better human-robot interaction and help humans to do many different and difficult tasks. Up to now, several methods have been implemented in robot imitation, which are devided in two categories: continuous and discrete. The continuous method is done without prior learning, which is mostly done on balancing the robot joints and increasing its speed. In the discrete method, the imitation is done by training the network in order to learn the robot in advance, and the focus is more on the prediction of movements by the robot to reduce the prediction time and increase the speed of the imitation. Despite the good results obtained in imitation, there are still challenges to its progress such as increasing the accuracy and speed of the robot in imitation, increasing the balance of the robot while walking, reducing the time to detect human movement in imitation. In this study, in order to imitation the robot of human movements, we first train the robot with the proposed neural network, CNN-LSTM neural network. To train the neural network, we use video images of the KARD data set, which we use in the preprocessing technique to improve the detection of human motion by the technique of optical flow, and also to avoid the problem of overfitting. In this study, we use the Nao robot to implement the proposed method. We add new gestures from the data set gestures using the interface of the Nao robot user interface to the gestures defined on the robot. Imitation motion by the Nao robot is performed discretely after predicting motion using network learning. The trained robot imitation the movements performed by humans, so we trained the robot to imitate in a discrete way. The results of the implementation of the proposed method include high accuracy and less prediction time and increase in the speed of robot imitation."

فایل: ّFile: Download فایل