基于Kinect的Nao机器人动作模仿系统的研究与实现

摘要

To avoid the complexity of the underlying motor control, make the robot realize motor skills through learning and improve its intelligence, Combining Somatosensory perception with humanoid robot Nao,this paper fo⁃cuses on the research of robot imitation learning,constructs the behavior imitation system and implements Nao robot's motion imitation using kinect based on the framework of imitation learning.By means of the skeleton tracking tech⁃nology of motion⁃sensing camera,the bone point information is collected,then be pre⁃processed into demonstration data.The demonstration data is encoded for representative learning through Gaussian mixture model (GMM),and then the output generalized by Gaussian mixture regression ( GMR) is mapped to Nao robot to realize the imitation of action.The experiment results indicate that Nao robot is able to implement behavior imitation in real⁃time and off⁃line mode and gives good effect owing to the fact that the motion trajectory is smooth and stable.%为避开复杂繁琐的底层运动控制，使机器人能够通过学习实现运动技能的获取，有效提高其智能性，将体态感知技术与仿人机器人Nao相结合，以机器人的模仿学习框架为指导，开发并实现了基于Kinect的Nao机器人动作模仿系统。利用Kinect体感摄像机的骨骼跟踪技术，采集示教者骨骼点信息，经预处理后得到示教数据，通过高斯混合模型（ GMM）对示教数据进行表征学习，经高斯混合回归（ GMR）泛化处理后，映射到Nao机器人中，实现动作的模仿。实验结果表明，Nao机器人能够进行实时和离线的动作模仿，运动轨迹平滑而稳定，动作模仿的效果较好。