Whole Body Control of Mobile Manipulators With Series Elastic Actuators for Cart Pushing Tasks.

Human-centered environments provide affordance for the use of two-handed mobile manipulators. Yet robots designed to function in and physically interact with such environments are not yet capable of meeting human users' requirements. This work proposes a whole body control framework of a two-handed mobile manipulator driven by series elastic actuators (SEAs) for cart pushing tasks. A whole body dynamic model for an integrated mobile platform and on-board arms is revealed, which takes into account the interaction forces with the cart. Then, the explicit force/position control of the mobile manipulator is performed. It enables the robot to interact dynamically with the environment while providing motion, i.e., the manipulators provide both output force control and motion control for pushing a cart. To cope with the highly nonlinear system dynamics and parameter variation of a SEA-driven mobile manipulator, this work proposes an adaptive robust controller based on a novel integral barrier Lyapunov function for cart pushing tasks by considering model uncertainty. The proposed controller enables the mobile manipulator to complete cart pushing tasks by regulating the position and output force of the mobile base and arms. The experimental results show the effectiveness of this approach in cart pushing tasks.

Complex-Valued Convolutional Gated Recurrent Neural Network for Ultrasound Beamforming.

Ultrasound detection is a potent tool for the clinical diagnosis of various diseases due to its real-time, convenient, and noninvasive qualities. Yet, existing ultrasound beamforming and related methods face a big challenge to improve both the quality and speed of imaging for the required clinical applications. The most notable characteristic of ultrasound signal data is its spatial and temporal features. Because most signals are complex-valued, directly processing them by using real-valued networks leads to phase distortion and inaccurate output. In this study, for the first time, we propose a complex-valued convolutional gated recurrent (CCGR) neural network to handle ultrasound analytic signals with the aforementioned properties. The complex-valued network operations proposed in this study improve the beamforming accuracy of complex-valued ultrasound signals over traditional real-valued methods. Further, the proposed deep integration of convolution and recurrent neural networks makes a great contribution to extracting rich and informative ultrasound signal features. Our experimental results reveal its outstanding imaging quality over existing state-of-the-art methods. More significantly, its ultrafast processing speed of only 0.07 s per image promises considerable clinical application potential. The code is available at https://github.com/zhangzm0128/CCGR.

Improving Classification Performance in Dendritic Neuron Models through Practical Initialization Strategies.

A dendritic neuron model (DNM) is a deep neural network model with a unique dendritic tree structure and activation function. Effective initialization of its model parameters is crucial for its learning performance. This work proposes a novel initialization method specifically designed to improve the performance of DNM in classifying high-dimensional data, notable for its simplicity, speed, and straightforward implementation. Extensive experiments on benchmark datasets show that the proposed method outperforms traditional and recent initialization methods, particularly in datasets consisting of high-dimensional data. In addition, valuable insights into the behavior of DNM during training and the impact of initialization on its learning performance are provided. This research contributes to the understanding of the initialization problem in deep learning and provides insights into the development of more effective initialization methods for other types of neural network models. The proposed initialization method can serve as a reference for future research on initialization techniques in deep learning.

MemeNet: Toward a Reliable Local Projection for Image Recognition via Semantic Featurization.

When we recognize images with the help of Artificial Neural Networks (ANNs), we often wonder how they make decisions. A widely accepted solution is to point out local features as decisive evidence. A question then arises: Can local features in the latent space of an ANN explain the model output to some extent? In this work, we propose a modularized framework named MemeNet that can construct a reliable surrogate from a Convolutional Neural Network (CNN) without changing its perception. Inspired by the idea of time series classification, this framework recognizes images in two steps. First, local representations named memes are extracted from the activation map of a CNN model. Then an image is transformed into a series of understandable features. Experimental results show that MemeNet can achieve accuracy comparable to most models' through a set of reliable features and a simple classifier. Thus, it is a promising interface to use the internal dynamics of CNN, which represents a novel approach to constructing reliable models.

Learning From Human Educational Wisdom: A Student-Centered Knowledge Distillation Method.

Existing studies on knowledge distillation typically focus on teacher-centered methods, in which the teacher network is trained according to its own standards before transferring the learned knowledge to a student one. However, due to differences in network structure between the teacher and the student, the knowledge learned by the former may not be desired by the latter. Inspired by human educational wisdom, this paper proposes a Student-Centered Distillation (SCD) method that enables the teacher network to adjust its knowledge transfer according to the student network's needs. We implemented SCD based on various human educational wisdom, e.g., the teacher network identified and learned the knowledge desired by the student network on the validation set, and then transferred it to the latter through the training set. To address the problems of current deficiency knowledge, hard sample learning and knowledge forgetting faced by a student network in the learning process, we introduce and improve Proportional-Integral-Derivative (PID) algorithms from automation fields to make them effective in identifying the current knowledge required by the student network. Furthermore, we propose a curriculum learning-based fuzzy strategy and apply it to the proposed PID control algorithm, such that the student network in SCD can actively pay attention to the learning of challenging samples after with certain knowledge. The overall performance of SCD is verified in multiple tasks by comparing it with state-of-the-art ones. Experimental results show that our student-centered distillation method outperforms existing teacher-centered ones.

A Herd-Foraging-Based Approach to Adaptive Coverage Path Planning in Dual Environments.

Coverage path planning (CPP) is essential for robotic tasks, such as environmental monitoring and terrain surveying, which require covering all surface areas of interest. As the pioneering approach to CPP, inspired by the concept of predation risk in predator-prey relations, the predator-prey CPP (PPCPP) has the benefit of adaptively covering arbitrary bent 2-D manifolds and can handle unexpected changes in an environment, such as the sudden introduction of dynamic obstacles. However, it can only work in bounded environment and cannot handle tasks in unbounded one, e.g., search and rescue tasks where the search boundary is unknown. Sometimes, robots are required to handle both bounded and unbounded environments, i.e., dual environments, such as capturing criminals in a city. Once encountering a building, the robot enters it to cover the bounded environment, then continues to cover the unbounded one when leaving the building. Therefore, the capability of swarm robots for the coverage tasks both in bounded and unbounded environments is important. In nature, herbivores live in groups to find more food and reduce the risk of predation. Especially the juvenile ones prefer to forage near the herd to protect themselves. Inspired by the foraging behavior of animals in a herd, this article proposes an online adaptive CPP approach that enables swarm robots to handle both bounded and unbounded environments without knowing the environmental information in advance, called dual-environmental herd-foraging-based CPP (DH-CPP). It's performance is evaluated in dual environments with stationary and dynamic obstacles of different shapes and quantity, and compared with three state-of-the-art approaches. Simulation results demonstrate that it is highly effective to handle dual environments.

Interpretability Diversity for Decision-Tree-Initialized Dendritic Neuron Model Ensemble.

To construct a strong classifier ensemble, base classifiers should be accurate and diverse. However, there is no uniform standard for the definition and measurement of diversity. This work proposes a learners' interpretability diversity (LID) to measure the diversity of interpretable machine learners. It then proposes a LID-based classifier ensemble. Such an ensemble concept is novel because: 1) interpretability is used as an important basis for diversity measurement and 2) before its training, the difference between two interpretable base learners can be measured. To verify the proposed method's effectiveness, we choose a decision-tree-initialized dendritic neuron model (DDNM) as a base learner for ensemble design. We apply it to seven benchmark datasets. The results show that the DDNM ensemble combined with LID obtains superior performance in terms of accuracy and computational efficiency compared to some popular classifier ensembles. A random-forest-initialized dendritic neuron model (RDNM) combined with LID is an outstanding representative of the DDNM ensemble.

A Robust Mean-Field Actor-Critic Reinforcement Learning Against Adversarial Perturbations on Agent States.

Multiagent deep reinforcement learning (DRL) makes optimal decisions dependent on system states observed by agents, but any uncertainty on the observations may mislead agents to take wrong actions. The mean-field actor-critic (MFAC) reinforcement learning is well-known in the multiagent field since it can effectively handle a scalability problem. However, it is sensitive to state perturbations that can significantly degrade the team rewards. This work proposes a Robust MFAC (RoMFAC) reinforcement learning that has two innovations: 1) a new objective function of training actors, composed of a policy gradient function that is related to the expected cumulative discount reward on sampled clean states and an action loss function that represents the difference between actions taken on clean and adversarial states and 2) a repetitive regularization of the action loss, ensuring the trained actors to obtain excellent performance. Furthermore, this work proposes a game model named a state-adversarial stochastic game (SASG). Despite the Nash equilibrium of SASG may not exist, adversarial perturbations to states in the RoMFAC are proven to be defensible based on SASG. Experimental results show that RoMFAC is robust against adversarial perturbations while maintaining its competitive performance in environments without perturbations.

Fully Complex-Valued Gated Recurrent Neural Network for Ultrasound Imaging.

Ultrasound imaging is widely used in medical diagnosis. It has the advantages of being performed in real time, cost-efficient, noninvasive, and nonionizing. The traditional delay-and-sum (DAS) beamformer has low resolution and contrast. Several adaptive beamformers (ABFs) have been proposed to improve them. Although they improve image quality, they incur high computation cost because of the dependence on data at the expense of real-time performance. Deep-learning methods have been successful in many areas. They train an ultrasound imaging model that can be used to quickly handle ultrasound signals and construct images. Real-valued radio-frequency signals are typically used to train a model, whereas complex-valued ultrasound signals with complex weights enable the fine-tuning of time delay for enhancing image quality. This work, for the first time, proposes a fully complex-valued gated recurrent neural network to train an ultrasound imaging model for improving ultrasound image quality. The model considers the time attributes of ultrasound signals and uses complete complex-number calculation. The model parameter and architecture are analyzed to select the best setup. The effectiveness of complex batch normalization is evaluated in training the model. The effect of analytic signals and complex weights is analyzed, and the results verify that analytic signals with complex weights enhance the model performance to reconstruct high-quality ultrasound images. The proposed model is finally compared with seven state-of-the-art methods. Experimental results reveal its great performance.

Learning Smooth Representation for Unsupervised Domain Adaptation.

Typical adversarial-training-based unsupervised domain adaptation (UDA) methods are vulnerable when the source and target datasets are highly complex or exhibit a large discrepancy between their data distributions. Recently, several Lipschitz-constraint-based methods have been explored. The satisfaction of Lipschitz continuity guarantees a remarkable performance on a target domain. However, they lack a mathematical analysis of why a Lipschitz constraint is beneficial to UDA and usually perform poorly on large-scale datasets. In this article, we take the principle of utilizing a Lipschitz constraint further by discussing how it affects the error bound of UDA. A connection between them is built, and an illustration of how Lipschitzness reduces the error bound is presented. A local smooth discrepancy is defined to measure the Lipschitzness of a target distribution in a pointwise way. When constructing a deep end-to-end model, to ensure the effectiveness and stability of UDA, three critical factors are considered in our proposed optimization strategy, i.e., the sample amount of a target domain, dimension, and batchsize of samples. Experimental results demonstrate that our model performs well on several standard benchmarks. Our ablation study shows that the sample amount of a target domain, the dimension, and batchsize of samples, indeed, greatly impact Lipschitz-constraint-based methods' ability to handle large-scale datasets. Code is available at https://github.com/CuthbertCai/SRDA.

Fully Complex-Valued Dendritic Neuron Model.

A single dendritic neuron model (DNM) that owns the nonlinear information processing ability of dendrites has been widely used for classification and prediction. Complex-valued neural networks that consist of a number of multiple/deep-layer McCulloch-Pitts neurons have achieved great successes so far since neural computing was utilized for signal processing. Yet no complex value representations appear in single neuron architectures. In this article, we first extend DNM from a real-value domain to a complex-valued one. Performance of complex-valued DNM (CDNM) is evaluated through a complex XOR problem, a non-minimum phase equalization problem, and a real-world wind prediction task. Also, a comparative analysis on a set of elementary transcendental functions as an activation function is implemented and preparatory experiments are carried out for determining hyperparameters. The experimental results indicate that the proposed CDNM significantly outperforms real-valued DNM, complex-valued multi-layer perceptron, and other complex-valued neuron models.

Solving Dynamic Traveling Salesman Problems With Deep Reinforcement Learning.

A traveling salesman problem (TSP) is a well-known NP-complete problem. Traditional TSP presumes that the locations of customers and the traveling time among customers are fixed and constant. In real-life cases, however, the traffic conditions and customer requests may change over time. To find the most economic route, the decisions can be made constantly upon the time-point when the salesman completes his service of each customer. This brings in a dynamic version of the traveling salesman problem (DTSP), which takes into account the information of real-time traffic and customer requests. DTSP can be extended to a dynamic pickup and delivery problem (DPDP). In this article, we ameliorate the attention model to make it possible to perceive environmental changes. A deep reinforcement learning algorithm is proposed to solve DTSP and DPDP instances with a size of up to 40 customers in 100 locations. Experiments show that our method can capture the dynamic changes and produce a highly satisfactory solution within a very short time. Compared with other baseline approaches, more than 5% improvements can be observed in many cases.

Extension of Stochastic Point Location for Multimodal Problems.

Stochastic point location (SPL) involves a learning mechanism (LM) determining an optimal point on the line when the only inputs LM receives are stochastic information about the direction in which LM should move. The complexity of SPL comes from the stochastic responses of the environment, which may lead LM completely astray. SPL is a fundamental problem in optimization and was studied by many researchers during the last two decades, including improvement of its solution and all-pervasive applications. However, all existing SPL studies assume that the whole search space contains only one optimal point. Since a multimodal optimization problem (MMOP) contains multiple optimal solutions, it is significant to develop SPL's multimodal version. This article extends it from a unimodal problem to a multimodal one and proposes a parallel partition search (PPS) solution to address this issue. The heart of the proposed solution involves extracting the feature of the historical sampling information to distinguish the subintervals that contain the optimal points or not. Specifically, it divides the whole search space into multiple subintervals and samples them parallelly, then utilizes the feature of the historical sampling information to adjust the subintervals adaptively and to find the subintervals containing the optimal points. Finally, the optimal points are located within these subintervals according to their respective sampling statistics. The proof of the ϵ -optimal property for the proposed solution is presented. The numerical testing results demonstrate the power of the scheme.

Evolutionary Weighted Broad Learning and Its Application to Fault Diagnosis in Self-Organizing Cellular Networks.

As a novel neural network-based learning framework, a broad learning system (BLS) has attracted much attention due to its excellent performance on regression and balanced classification problems. However, it is found to be unsuitable for imbalanced data classification problems because it treats each class in an imbalanced dataset equally. To address this issue, this work proposes a weighted BLS (WBLS) in which the weight assigned to each class depends on the number of samples in it. In order to further boost its classification performance, an improved differential evolution algorithm is proposed to automatically optimize its parameters, including the ones in BLS and newly generated weights. We first optimize the parameters with a training dataset, and then apply them to WBLS on a test dataset. The experiments on 20 imbalanced classification problems have shown that our proposed method can achieve higher classification accuracy than the other methods in terms of several widely used performance metrics. Finally, it is applied to fault diagnosis in self-organizing cellular networks to further show its applicability to industrial application problems.

Domain Adaptation Multitask Optimization.

Multitask optimization (MTO) is a new optimization paradigm that leverages useful information contained in multiple tasks to help solve each other. It attracts increasing attention in recent years and gains significant performance improvements. However, the solutions of distinct tasks usually obey different distributions. To avoid that individuals after intertask learning are not suitable for the original task due to the distribution differences and even impede overall solution efficiency, we propose a novel multitask evolutionary framework that enables knowledge aggregation and online learning among distinct tasks to solve MTO problems. Our proposal designs a domain adaptation-based mapping strategy to reduce the difference across solution domains and find more genetic traits to improve the effectiveness of information interactions. To further improve the algorithm performance, we propose a smart way to divide initial population into different subpopulations and choose suitable individuals to learn. By ranking individuals in target subpopulation, worse-performing individuals can learn from other tasks. The significant advantage of our proposed paradigm over the state of the art is verified via a series of MTO benchmark studies.

LAGAM: A Length-Adaptive Genetic Algorithm With Markov Blanket for High-Dimensional Feature Selection in Classification.

Feature selection (FS) is an essential technique widely applied in data mining. Recent studies have shown that evolutionary computing (EC) is very promising for FS due to its powerful search capability. However, most existing EC-based FS methods use a length-fixed encoding to represent feature subsets. This inflexible encoding turns ineffective when high-dimension data are handled, because it results in a huge search space, as well as a large amount of training time and memory overhead. In this article, we propose a length-adaptive genetic algorithm with Markov blanket (LAGAM), which adopts a length-variable individual encoding and enables individuals to evolve in their own search space. In LAGAM, features are rearranged decreasingly based on their relevance, and an adaptive length changing operator is introduced, which extends or shortens an individual to guide it to explore in a better search space. Local search based on Markov blanket (MB) is embedded to further improve individuals. Experiments are conducted on 12 high-dimensional datasets and results reveal that LAGAM performs better than existing methods. Specifically, it achieves a higher classification accuracy by using fewer features.

Learning-Based Cuckoo Search Algorithm to Schedule a Flexible Job Shop With Sequencing Flexibility.

This work considers an extended version of flexible job-shop problem from a postprinting or semiconductor manufacturing environment, which needs a directed acyclic graph rather than a linear order to describe the precedences among operations. To obtain its reliable and high-quality schedule in a reasonable time, a learning-based cuckoo search (LCS) algorithm is presented. In it, cuckoo search is selected as an optimizer. To produce promising solutions in a high-dimensional solution space, a sparse autoencoder is introduced to compress a high-dimensional solution into an informative low-dimensional one. It extends the application area of autoencoder-embedded evolutionary optimization methods into combinational optimization by developing an improved one-hot encoding method. Then, in order to reveal the linkages among decision variables and enhance the explore ability of the proposed method, a factorization machine (FM) is used, for the first time, to capture the relevant and complementary features of population. Hence, a parallel framework involving three co-evolved subpopulations is constructed. The first one is an autoencoder embedded subpopulation, the second one is assisted by an FM, and the last one undergoes a regular iteration process. To balance the exploration and exploitation of the proposed framework and avoid unnecessary computation, a reinforcement learning algorithm is used to adaptively adjust the proportion of subpopulations and tune parameters of each subpopulation iteratively. Numerical simulations with benchmarks are performed to compare it with CPLEX, some classical heuristics, and several recently developed methods. The results shows that it well outperforms them.

Information-Theory-based Nondominated Sorting Ant Colony Optimization for Multiobjective Feature Selection in Classification.

Feature selection (FS) has received significant attention since the use of a well-selected subset of features may achieve better classification performance than that of full features in many real-world applications. It can be considered as a multiobjective optimization consisting of two objectives: 1) minimizing the number of selected features and 2) maximizing classification performance. Ant colony optimization (ACO) has shown its effectiveness in FS due to its problem-guided search operator and flexible graph representation. However, there lacks an effective ACO-based approach for multiobjective FS to handle the problematic characteristics originated from the feature interactions and highly discontinuous Pareto fronts. This article presents an Information-theory-based Nondominated Sorting ACO (called INSA) to solve the aforementioned difficulties. First, the probabilistic function in ACO is modified based on the information theory to identify the importance of features; second, a new ACO strategy is designed to construct solutions; and third, a novel pheromone updating strategy is devised to ensure the high diversity of tradeoff solutions. INSA's performance is compared with four machine-learning-based methods, four representative single-objective evolutionary algorithms, and six state-of-the-art multiobjective ones on 13 benchmark classification datasets, which consist of both low and high-dimensional samples. The empirical results verify that INSA is able to obtain solutions with better classification performance using features whose count is similar to or less than those obtained by its peers.

Comparative Study on Noise-Estimation-Based Fuzzy C-Means Clustering for Image Segmentation.

Since a noisy image has inferior characteristics, the direct use of Fuzzy C-Means (FCM) to segment it often produces poor image segmentation results. Intuitively, using its ideal value (noise-free image) benefits FCM's robustness enhancement. Therefore, the realization of accurate noise estimation in FCM is a new and important task. To date, only two noise-estimation-based FCM algorithms have been proposed for image segmentation, that is: 1) deviation-sparse FCM (DSFCM) and 2) our earlier proposed residual-driven FCM (RFCM). In this article, we make a thorough comparative study of DSFCM and RFCM. We demonstrate that an RFCM framework can realize more accurate noise estimation than DSFCM when different types of noise are involved. It is mainly thanks to its utilization of noise distribution characteristics instead of noise sparsity used in DSFCM. We show that DSFCM is a particular case of RFCM, thus signifying that they are the same when only impulse noise is involved. With a spatial information constraint, we demonstrate RFCM's superior effectiveness and efficiency over DSFCM in terms of supporting experiments with different levels of single, mixed, and unknown noise.

Divergent Component of Motion Planning and Adaptive Repetitive Control for Wearable Walking Exoskeletons.

Wearable walking exoskeletons show great potentials in helping patients with neuro musculoskeletal stroke. Key to the successful applications is the design of effective walking trajectories that enable smooth walking for exoskeletons. This work proposes a walking planning method based on the divergent component of motion to obtain a stable joint angle trajectory. Since periodic and nonperiodic disturbances are ubiquitous in the repeating walking motion of an exoskeleton system, a major challenge in the walking control of wearable exoskeleton is the joint angle drift problem, that is, the joint angle motion trajectories are not necessarily periodic due to the presence of disturbance. To address this challenge, this work develops an adaptive repetitive control strategy to guarantee that the motion trajectories of joint angle are repetitive. In particular, by treating the disturbance as system uncertainties, an adaptive controller is designed to compensate for the uncertainties based on an integral-type Lyapunov function. A fully saturated learning approach is then developed to achieve asymptotic tracking of repetitive walking trajectories. Extensive experiments are carried out to demonstrate the effectiveness of the tracking performance.