Yuan Wei

In this paper, the human-like motion issue for anthropomorphic arms is further discussed. An Intelligent Human-like Motion Planner (IHMP) consisting of Movement Primitive (MP), Bayesian Network (BN) and Coupling Neural Network (CPNN) is proposed to help the robot generate human-like arm movements. Firstly, the arm motion model is decoupled in the aspects of arm structure and motion process, respectively. In the former aspect, the arm model is decoupled into different simple models through the Movement Primitive. A Hierarchical Planning Strategy (HPS) is proposed to decouple a complete motion process into different sub-processes. Based on diversified arm motion models, the Bayesian Network is used to help the robot choose the suitable motion model among these arm motion models. Then, according to the features of diversified arm motion models, the Coupling Neural Network is proposed to obtain the inverse kinematic (IK) solutions. This network can integrate different models into a single network and reflect the features of these models by changing the network structure. Being a major contribution to this paper, specific focus is on the improvement of human-like motion accuracy and independent consciousness of robots. Finally, the availability of the IHMP is verified by experiments on a humanoid robot Pepper.

As a single-layer feedforward network (SLFN), extreme learning machine (ELM) has been successfully applied for classification and regression in machine learning due to its faster training speed and better generalization. However, it will perform poorly for domain adaptation in which the distributions between training data and testing data are inconsistent. In this article, we propose a novel ELM called two-stage transfer extreme learning machine (TSTELM) to solve this problem. At the statistical matching stage, we adopt maximum mean discrepancy (MMD) to narrow the distribution difference of the output layer between domains. In addition, at the subspace alignment stage, we align the source and target model parameters, design target cross-domain mean approximation, and add the output weight approximation to further promote the knowledge transferring across domains. Moreover, the prediction of test sample is jointly determined by the ELM parameters generated at the two stages. Finally, we investigate the proposed approach in classification task and conduct experiments on four public domain adaptation datasets. The result indicates that TSTELM could effectively enhance the knowledge transfer ability of ELM with higher accuracy than other existing transfer and non-transfer classifiers.

Extreme Learning Machine (ELM) is widely used in various fields because of its fast training and high accuracy. However, it does not primarily work well for Domain Adaptation (DA) in which there are many annotated data from auxiliary domain and few even no annotated data in target domain. In this paper, we propose a new variant of ELM called Discriminative Extreme Learning Machine with Cross‐Domain Mean Approximation (DELM‐CDMA) for unsupervised domain adaptation. It introduces Cross‐Domain Mean Approximation (CDMA) into the hidden layer of ELM to reduce distribution discrepancy between domains for domain bias elimination, which is conducive to train a high accuracy ELM on annotated data from auxiliary domains for target tasks. Linear Discriminative Analysis (LDA) is also adopted to improve the discrimination of learned model and obtain higher accuracy. Moreover, we further provide a Discriminative Kernel Extreme Learning Machine with Cross‐Domain Mean Approximation (DKELM‐CDMA) as the kernelization extension of DELM‐CDMA. Some experiments are performed to investigate the proposed approach, and the result shows that DELM‐CDMA and DKELM‐CDMA could effectively extend ELM suitable for domain adaptation and outperform ELM and many other domain adaptation approaches.

Humanoid robot is one of the most active frontiers in robotics. As the basic part of motion of humanoid robot, human-like motion planning is always one of the research hotspots and difficulties. In this paper, a comprehensive approach is proposed to help robot NAO generate the human-like arm movements. Firstly, a novel arm motion mode is built based on the Movement Primitives (MPs). The whole arm movement can be decoupled into different simple sub-movements with different motion models, which improves the accuracy and computational efficiency of human-like arm movements. Then a motion decision algorithm based on Bayesian Network (BN) is proposed to help robot NAO choose the suitable motion model among these MPs. Finally, according to the structure features of the MPs, the IK solutions can be classified into two categories: methods based on index and on geometrical constraints. Through the comprehensive approach, the robot NAO can generate various human-like arm movements with satisfactory accuracy. The availability of the approach is verified by similarity experiment and human-like movement experiment.

SUMMARY: Human-like motion of robots can improve human–robot interaction and increase the efficiency. In this paper, a novel human-like motion planning strategy is proposed to help anthropomorphic arms generate human-like movements accurately. The strategy consists of three parts: movement primitives, Bayesian network (BN), and a novel coupling neural network (CPNN). The movement primitives are used to decouple the human arm movements. The classification of arm movements improves the accuracy of human-like movements. The motion-decision algorithm based on BN is able to predict occurrence probabilities of the motions and choose appropriate mode of motion. Then, a novel CPNN is proposed to solve the inverse kinematics problems of anthropomorphic arms. The CPNN integrates different models into a single network and reflects the features of these models by changing the network structure. Through the strategy, the anthropomorphic arms can generate various human-like movements with satisfactory accuracy. Finally, the availability of the proposed strategy is verified by simulations for the general motion of humanoid NAO.

Purpose: This paper aims to deal with the problem of designing robot behaviors (mainly to robotic arms) to express emotions. The authors study the effects of robot behaviors from our humanoid robot NAO on the subject’s emotion expression in human–robot interaction (HRI). Design/methodology/approach: A method to design robot behavior through the movement primitives is proposed. Then, a novel dimensional affective model is built. Finally, the concept of action semantics is adopted to combine the robot behaviors with emotion expression. Findings: For the evaluation of this combination, the authors assess positive (excited and happy) and negative (frightened and sad) emotional patterns on 20 subjects which are divided into two groups (whether they were familiar with robots). The results show that the recognition of the different emotion patterns does not have differences between the two groups and the subjects could recognize the robot behaviors with emotions. Practical implications: Using affective models to guide robots’ behavior or express their intentions is highly beneficial in human–robot interaction. The authors think about several applications of the emotional motion: improve efficiency in HRI, direct people during disasters, better understanding with human partners or help people perform their tasks better. Originality/value: This paper presents a method to design robot behaviors with emotion expression. Meanwhile, a similar methodology can be used in other parts (leg, torso, head and so on) of humanoid robots or non-humanoid robots, such as industrial robots.