Teleoperation is one of the most efficient ways to control a robot remotely. It reproduces the movements of the robot as a human exhibits in real time. In this paper, a framework is proposed for a Kinect-based teleoperation which allows the NAO humanoid robot to imitate the recognised human motions using Hidden Markov Model (HMM). Microsoft Kinect v2.0 is used to capture the human motions, which makes the framework more user-friendly. The skeleton joint positions obtained from Kinect are processed to generate the robot’s joint angles using simple analytical geometry and vector algebra methods. Data collection of all kinds of motions that human can do is a challenging and time-consuming process in humanoid robot teleoperation. Therefore, a data augmentation method is introduced and applied to the joint angles mapped from the human skeleton data. It increases the data artificially without actually collecting new data and performs well to train the model. A promising result as compared to state-of-the-art techniques is achieved with our proposed method for teleoperating a NAO humanoid robot.

远程操作是远程控制机器人的最有效方法之一。它实时再现机器人的动作，就像人类展示的一样。在本文中，提出了一种基于 Kinect 的遥操作框架，该框架允许 NAO 仿人机器人使用隐马尔可夫模型 (HMM) 模仿已识别的人体动作。使用 Microsoft Kinect v2.0 捕捉人体动作，使框架更加人性化。使用简单的解析几何和矢量代数方法处理从 Kinect 获得的骨架关节位置以生成机器人的关节角度。人类可以做的各种运动的数据收集是仿人机器人遥操作中一个具有挑战性和耗时的过程。因此，引入了一种数据增强方法并将其应用于从人体骨骼数据映射的关节角度。它在没有实际收集新数据的情况下人为地增加了数据，并且在训练模型方面表现良好。与最先进的技术相比，我们提出的用于远程操作 NAO 人形机器人的方法取得了有希望的结果。