Font Size: a A A

Dance Planning For Entertainment Robot And Computational Implementation

Posted on:2021-01-17Degree:DoctorType:Dissertation
Country:ChinaCandidate:R Q WuFull Text:PDF
GTID:1525306323974669Subject:Intelligent Science and Technology
Abstract/Summary:PDF Full Text Request
Combined with development of robot technology,music and dance,robot dance planning is an emerging research field.As a multidisciplinary research field,the problems to be solved by robot dance planning and the problems of other disciplines can inspire and promote each other in terms of ideas and research tools.Simultaneously,robot dance also plays an active role in the inheritance and spread of dance culture.In addition,robot dance also has a very wide application potential in home entertainment,emotional escort and other application scenarios.The core issues of robot dance planning can be attributed to aesthetic evaluation for robot dance,motion balance control and motion planning.It is the goal of this field that how to plan a dance that meets the aesthetic and environmental background under the premise of ensuring the balance of the robot.Therefore,this article takes the robot dance planning based on music as the research goal,and have done a series of studies on these core problems.The main research work of this paper are as follows:1.Before studying the core issues of robot dance planning,a series of preparatory research work has been done.First,we design an experimental platform based on an entertainment robot.We have been developed a robot control system to make the AlphalP robot available for scientific research,and produced a series of robot dance data sets for different research objects.Then,in order to facilitate the scientific computation of robot dance,a series of basic processing methods for robot dance was proposed.These methods include:pose similarity computation method,dance similarity computation method and dance scaling algorithm.These preparations are necessary prerequisites for follow-up research work.To promote the development of robot dance planning,these works have been open sourced.2.Due to the subjectivity of artistic aesthetics,it is difficult to establish quantitative stan dards directly for the aesthetic evaluation of robot dance.For this problem,this paper proposes a Wasserstein GAN based model and a SinGAN based model to solve the aesthetic evaluation in model training.After the model has been trained,the discriminator of the WGAN-gp model can be used for classification tasks of dancing style.While solving the evaluation problem,the generator of the models can also be used to generate dance materials.In the models of this paper proposed,the generators are able to generate dance pose and dance motion sequences of the same style,respectively.3.In the current robot dance research,there are few researches on motion balance.In this research works of motion balance,most of their research uses ZMP balance control theory for advanced robot.However,research on the balance control of robots without corresponding sensors is still a blank.For the motion balance problem of cheap robots,we design a balance control model based on deep reinforcement learning and sample augmentation technology.The proposed model can be trained on the physical robot directly.After the model has been trained,according to the current posture of the robot,the Actor network of the model can generate balancing actions continuously.And,the Critic network can be used to determine whether the actions are balanced.4.In the existing planning methods,constructing a motion library is mostly used to avoid the motion balance problem in robot dance planning,which also limits the diversity of dance.Combined with the previous work,this paper proposes a dance planning model that considers the motions balance.In this model,the music information is converted into a dance feature sequence through a feature matching model firstly.Then,the dance planning algorithm generates a dance motion sequence based on the dance feature sequence.The processing steps of the planning algorithm are as follows:First,the dance poses data set is expanded by the dance action sequence which generated by the SinGAN model.Secondly,according to the dance features,the dance actions are combined from the dance posture data set,and the actions are filtered by the balance judgment module.Finally,based on all the dance features,a dance motion sequence is planned from the filtered actions.With the advantages of the poseture augmentation module and the balanced action filter module,the model can break through the limitations of the action library and generate a large number of dance motions that meet the dance features.This feature of the model also enhances the diversity of the generated dance.The actual effects of the above work have been verified in experiments with the help of related tools and data sets.The designed cheap robot experimental platform performed well in several research tasks and fully meet the needs of scientific research.Moreover,the open-source toolkit lowers the barriers to entry in the field of robot dance research,and it allows researchers to focus more on the core issues of robotic dance research.The work using GAN model for the aesthetic evaluation problem has solved the evaluation problem in the training of the dance planning model,and it also verifies the feasibility of using GAN model to solve the robot dance problem.The balance control work based on deep reinforcement learning fills the blank in the study of motion balance in sensorless robot,and the motion balance judgement model can also be applied to the other research work.The proposed dance planning algorithm can plan a highly innovative dance based on music.This planning algorithm combines the results of the previous work,and it also proves the practicality of the previous work from the side.
Keywords/Search Tags:Intelligent Robot, Robot Dance, Motion Planning, Dancing Posture Generation, Robot Balance Control
PDF Full Text Request
Related items