Deficit in imitation skills is one of the core symptoms of children with Autism Spectrum Disorder (ASD). In this study, we have tried to look closer at the body gesture imitation performance of 20 participants with autism, i.e. ASD group, and 20 typically developing subjects, i.e. TD group, in a set of robot-child and human-child gross imitation tasks. The results of manual scoring by two specialists indicated that while the TD group showed a significantly better imitation performance than the ASD group during the tasks, both ASD and TD groups performed better in the human-child mode than the robot-child mode in our experimental setup. Next, to introduce an automated imitation assessment system, we present different mathematical models of the children’s imitation performance using some State-Image based algorithms including Acceptable Bound, Mahalanobis Distance, and Signals’ Cross-Correlations as well as Hidden Markov Models based on the time-dependent kinematics data of the participants’ joints. Among the different studied models, we observed that the “State-Image Acceptable Bound method with position, velocity, and acceleration features” is the best one. This method has a mean Pearson correlation of ~ 45%, which is fairly comparable to the related works (out of autism field) in assessing the quality of dynamic actions. Finally, for a treatment application of using artificial intelligence algorithms in automated evaluation of children’s behaviors as an unbiased and quantifiable measurement in HRI, we propose a reciprocal gross imitation human–robot interaction platform with the potential to aid in the cognitive rehabilitation of children with autism.

模仿能力不足是自闭症谱系障碍（ASD）儿童的核心症状之一。在这项研究中，我们试图仔细观察一组机器人-儿童和人类-儿童的总体模仿任务中，自闭症参与者（即ASD组）和20个典型发展中的受试者（即TD组）的手势模仿性能。两位专家的人工评分结果表明，尽管TD组在任务执行过程中的模仿性能明显优于ASD组，但在我们的实验中，ASD和TD组在儿童模式下的表现均优于机器人儿童模式建立。接下来，为了介绍自动模仿评估系统，我们使用一些基于状态图像的算法（包括“可接受范围”，“ 马哈拉诺比斯距离和信号的互相关性以及基于参与者关节随时间变化的运动学数据的隐马尔可夫模型。在不同的研究模型中，我们观察到“具有位置，速度和加速度特征的状态图像可接受约束方法”是最好的方法。这种方法的平均皮尔逊相关系数约为45％，与评估动态动作质量的相关工作（自闭症领域之外）相当。最后，对于使用人工智能算法自动评估儿童行为作为HRI中无偏且可量化的测量方法的治疗应用，我们提出了一种互惠的总体模仿人机交互平台，该平台有可能有助于自闭症儿童的认知康复。信号的互相关性以及基于参与者关节随时间变化的运动学数据的隐马尔可夫模型。在不同的研究模型中，我们观察到“具有位置，速度和加速度特征的状态图像可接受约束方法”是最好的方法。这种方法的平均皮尔逊相关系数约为45％，与评估动态动作质量的相关工作（自闭症领域之外）相当。最后，对于使用人工智能算法自动评估儿童行为作为HRI中无偏且可量化的测量方法的治疗应用，我们提出了一种互惠的总体模仿人机交互平台，该平台有可能有助于自闭症儿童的认知康复。信号的互相关性以及基于参与者关节随时间变化的运动学数据的隐马尔可夫模型。在不同的研究模型中，我们观察到“具有位置，速度和加速度特征的状态图像可接受约束方法”是最好的方法。这种方法的平均皮尔逊相关系数约为45％，与评估动态动作质量的相关工作（自闭症领域之外）相当。最后，对于使用人工智能算法自动评估儿童行为作为HRI中无偏且可量化的测量方法的治疗应用，我们提出了一种互惠的总体模仿人机交互平台，该平台有可能有助于自闭症儿童的认知康复。