Artificial Life and Robotics最新文献

This research introduces the novel “Kraken-Gear” mechanism, emphasizing the advantages of additive polymer 3D printing in high-ratio gearbox systems for lightweight robotic applications, such as surgical instruments. The innovative kinematic solution provides high torsional system stiffness, substantial gear ratios, and backlash-free transmission. Leveraging the “hot lithography” additive manufacturing method ensures precise and warp-free gearbox components. Targeting medical technology, the gearbox meets stringent requirements: backlash-free, minimal vibration, high precision, and torque, with minimized weight for ergonomic comfort and fatigue mitigation. Computational simulations assess forces and stresses, highlighting the potential of additive manufacturing for cost-effective and functionally efficient gearbox fabrication. Nevertheless, careful material selection remains imperative for optimal functionality, especially in demanding medical applications. In summary, this research underscores a promising approach to gearbox fabrication, emphasizing the critical role of material selection and simulation-based assessments for optimal performance.

Thermal skin images are used to evaluate the physiological and psychological states of patients. To implement remote daily health monitoring, we attempted to assess subjective health conditions using thermal face images. In our previous study, we constructed an anomaly detection model to detect poor health conditions; the area under the receiver operating characteristic curve of the anomaly-detection model was 0.70. However, how the spatial distribution of facial skin temperature changes in response to subjective health conditions remains unclear. In this study, we statistically analyzed the acquired thermal face images to investigate the effect of subjective health conditions on facial skin temperature distribution. As a result of the comparison between health conditions, we confirmed that typically the average temperatures and left-right asymmetry in some regions of the face were significantly higher in poor health conditions than in good health conditions.

Reinforcement learning is a mathematical framework for learning better choices through trial-and-error. Recent studies revealed that reinforcement learning is applicable to animal behavior and cognition. However, applying reinforcement learning to animal behavior sometimes encounters difficulties because the information sources utilized by animals to make choices are often unknown, whereas this is identified as the “state” in the reinforcement learning framework. We sought to identify possible state settings including non-standard formulations suitable for explaining data from past chimpanzee studies. Although chimpanzees’ performance in a serial learning task was inconsistent with standard reinforcement learning formulations, we found that the combination of state-independent choice making and state-dependent evaluation produced consistent results. Exploration of state settings in reinforcement learning may shed new light on animal learning processes.

With the growing population of English learners, how to improve the efficiency of English learning has become a focus of research. This article focuses on automatic error-checking in English short text translation. The Transformer model was enhanced by combining with the bidirectional gated recurrent unit (BiGRU) algorithm to create a dual-encoder model that better captures information within input sequences. Experiments were then conducted on different corpora. The improved Transformer model obtained a ({text{F}}_{0.5}) of 59.09 on CoNLL-2014 and 61.05 Google-bilingual evaluation understudy (GLEU) on JFLEG, both of which were better than the other methods compared. The case analysis showed that the improved Transformer model accurately found errors in short text translation. The findings indicate that the proposed approach is reliable in the automatic error-checking of English short text translation and can be applied in practice.

Real-time English speech translation is useful in numerous situations, including business and travel. The goal of this research is to improve real-time English speech translation efficacy. Initially, filter bank (FBank) features were extracted from English speech. Subsequently, an enhanced Transformer model was introduced, incorporating a causal convolution module in the front end of the encoder to capture English speech features with location information. The performance of the optimized model in translating English speech to different target languages was tested using the MuST-C dataset. The results revealed differences in translation results for different target languages using the improved Transformer. The highest bilingual evaluation understudy (BLEU) score was observed for Spanish text at 20.84, while Russian text obtained the lowest score of 10.56. The average BLEU score was 18.51, with an average lag time delay of 1202.33 ms. Compared to the conventional Transformer model, the improved model exhibited higher BLEU scores, lower time delay, and optimal performance when utilizing a convolutional kernel size of 3 × 3. The results demonstrate the dependability of the improved Transformer model in real-time English speech translation, highlighting its practical usefulness.

Gait-type recognition is important for robotic exoskeletons and walking-assistance devices to adjust their output according to the users’ needs. However, the growing trend of using machine learning (ML) models is both labor- and data-intensive, which makes it practically less attractive for application in exoskeletons and wearable-assistive devices. This research aims to devise a fuzzy-based gait recognition algorithm that requires minimum training data (only 40 cycles for each of the 5 gait types) and adapts to new users without having the need of pre-training for each of them. The proposed algorithm uses the fuzzy logic system (FLS) and Welford’s (variance computation) method to enhance the adaptability by adjusting the rules for gait-type recognition and fine-tuning them in real time for every new user without requiring a specific prior training. Simulation-based evaluation of the proposed algorithm shows a gait-type recognition accuracy of 63.0%, an improvement of 36.8% over the non-adaptive fuzzy-based recognition algorithm. Moreover, the results show that the proposed algorithm outperforms the popular ML methods (support vector machine, Naive Bayes classifier, and logistic regression) when subjected to limited gait-cycles data and no prior training is provided.

In the near future, autonomous vehicles will be able to share information with other vehicles via communication, enabling appropriate traffic control approaches. A new approach to traffic control using fully autonomous driving involves the realization of a flat interchange. Unlike conventional approaches, this study focused on roads and interchanges that do not assume lanes. Specifically, we propose an interchange flow control approach to traffic control using a virtual wall (VW), which acquires and shares the initial position, destination, and speed of all vehicles entering an interchange in a two-dimensional space where vehicles can move freely, and then realizes appropriate control based on this information. Each vehicle individually calculates the shortest path to avoid the VW, thereby realizing a safe and rational path selection. In this study, a genetic algorithm was used to determine the location of the VW. The effectiveness of the proposed method was evaluated using simulations, and the results showed that compared to manual deployment in the roundabout form, the proposed method using VWs reduced the total path length and the number of collisions to zero. In addition, when comparing the case where VWs were deployed in common for all vehicles and the case where VWs were deployed individually for each vehicle, it was shown that the total path length was shorter when individual VWs were deployed.

Character-level convolutional neural networks (CLCNNs) are commonly used to classify textual data. CLCNN is used as a more versatile tool. For natural language recognition, after decomposing a sentence into character units, each unit is converted into a corresponding character code (e.g., Unicode values) and the code is input into the CLCNN network. Thus, sentences can be treated like images. We have previously applied a CLCNN to verify whether a university’s diploma and/or curriculum policies are well written. In this study, we experimentally confirm the effectiveness of CLCNN using tweet data. In particular, we focus on the effect of the number of units on performance using the following two types of data; one is a real and public tweet dataset on the reputation of a cell phone, and the other is the NTCIR-13 MedWeb task, which consists of pseudo-tweet data and is a well-known collection of tests for multi-label problems. Results of experiments conducted by varying the number of units in the all-coupled layer confirmed the agreement of the results with the theorem introduced in the Amari’s book (Amari in Mathematical Science New Development of Information Geometry, For Senior & Graduate Courses. SAIENSU-SHA Co., 2014). Furthermore, in the NTCIR-13 MedWeb task, we analyze two kinds of experiments, the effects of kernel size and weight perturbation. The results of the difference in the kernel size suggest the existence of an optimal kernel size for sentence comprehension. The results of perturbations to the convolutional layer and pooling layer indicate the possibility of relationship between the numbers of degrees of freedom and network parameters.

In recent years, many studies have used Convolutional Neural Networks (CNN) as an approach to automate bin picking tasks by robots. In these previous studies, all types of objects in a bin were used as training data to optimize CNN parameters. Therefore, as the number of types of objects in a bin increases under the condition that the total number of training data is retained, CNN is not sufficiently optimized. In this study, we propose a learning method of CNN to achieve a bin picking task for multi-types of objects. Unlike previous learning method using multi-types of objects, we use a single type of object with a well-designed shape to obtain training data. It is true that when an object is used for training, then the learning for the corresponding object proceeds very well. However, the training does not contribute so much to the leaning of other types of objects. we expect the CNN to learn many types of grasping methods simultaneously by the training data of the single well-designed object. In that case, the total number of training data for each type of objects in a bin can be retained even if the number of types of objects increases. To verify the idea, we construct 12 different CNN models which are trained by different types of objects. Through simulations and robot experiments, bin picking tasks to pick multi-types of objects were performed using those CNN models. As a result, the training method which uses a complex-shaped object achieved higher grasping success rate than the training method which uses a primitive-shaped object. Moreover, the training method which uses a complex-shaped object achieved higher grasping success rate than the previous training method which uses all types of objects in a bin.

This paper proposes a simple method based on a novel use of elitism to increase the population size of artificial creatures while minimizing evaluation cost. This can contribute to preventing premature convergence of the population. We propose the “Excessive Elitism (EE)” method by modifying elitism in HyperNEAT (Hypercube-based NeuroEvolution of Augmenting Topologies), which is an evolutionary algorithm commonly used to evolve genotype [i.e., Compositional Pattern Producing Network (CPPN)] of artificial creatures. In EE, the evaluated fitness of best-fit individuals will be succeeded and reused instead of being re-evaluated during subsequent fitness evaluations, thereby reducing the evaluation cost if the elite size is excessive. Notably, EE also disables speciation and fitness sharing, serving to simplify the population structure and reduce complexity. In a 3D multi-agent environment, we evolved the morphology and behavior of artificial creatures with a simple target approach task. We assumed a baseline case (EE (2, 20)) in which a small population size was used due to the strong limitation of the evaluation cost and adopted a normal small elite size. This often led to premature convergence of the population to suboptimal individuals who could not reach the target. However, with the application of EE, the population was capable of evolving to reach the target, maintaining an evaluation cost comparable to EE (2, 20). We demonstrate that EE method serves as a simpler alternative to speciation for diversity preservation, capable of enhancing both the average and optimal fitness of a population, thus preventing premature convergence at a minimal evaluation cost. Further research in complex environments is required to fully uncover the potential and limitations of this method.