Artificial Life and Robotics最新文献

This study aims to develop a smart fish feeding robot integrated with a water level control and an Internet of Things (IoT)-based monitoring system for aquaculture ponds. The proposed system comprises several key components, namely, a feed storage tank, a feed distribution mechanism, a feed ejection mechanism, an ejector position turning mechanism, a water level control system, a structural base, and a solar-powered energy supply. Three DC motors serve as actuators for the mechanical subsystems, while ultrasonic sensors are deployed to measure both feed quantity and water levels. The water level control operates through two water pumps, one for draining excess water when the level surpasses a predefined maximum threshold, and the other for filling when the level drops below a minimum threshold. The IoT-based system facilitates real-time control and monitoring of both feeding and water conditions. The control algorithms comprise On–Off and Proportional-Integral-Derivative (PID) controllers for actuator regulation, whereas the water pumps are controlled using a Proportional-Integral (PI) controller. The PID controller is particularly employed for the feed ejection mechanism to maintain precise angular velocity and prevent feed degradation, whereas On–Off controller suffices for the other two mechanisms due to their lower precision requirements. Experimental evaluations show consistent feeding performance, with mass of ejected feed remained consistently within a narrow range of 500–510 [g] for the setting of 500 [g] and 1000–1010 [g] for the setting of 1000 [g], while the ejection longest distance remained within the range of 6.4–6.8 [m], across, respectively, ten ejections experiment. The ejection motor reached the desired speed of 1600 [rpm] in approximately 1.5 [s]. The water level control system successfully achieved the target level of 32 [cm] in approximately 45 [s]. These findings confirm the reliability and stability of the developed robotic system for sustainable aquaculture practices.

Nailfold capillaries, which are located at the base of the fingernail, provide a non-invasive window into peripheral microcirculation and are valuable for the early detection of systemic diseases such as diabetes mellitus and scleroderma, with abnormalities in capillary morphology, such as twisting and enlargement, indicative of vascular damage due to factors such as glucose accumulation. Traditional imaging modalities such as ultrasound and laser imaging have been used to evaluate peripheral blood vessels; however, such methods are often expensive and burdensome. Recently, the use of machine learning with microscopic imaging has shown promise in classifying capillary structures. However, these methods typically require large, annotated datasets, limiting the practicality; thus, we propose a novel approach that uses transfer learning to classify the shape of nailfold capillaries, specifically the presence or absence of capillary twisting. A pre-trained model based on fundus vessel images was fine-tuned for this task, leveraging the structural similarity between retinal and nailfold capillaries. The proposed method was found to achieve accurate and efficient classification even with a limited nailfold dataset, demonstrating its potential for practical clinical use.

This study proposes a compact, low-cost near-infrared fundus imaging system that addresses the limitations of conventional large and expensive devices, enabling non-invasive and non-mydriatic retinal evaluation. Retinal videos were acquired using a monochrome camera, followed by motion correction based on template matching, and vessel enhancement using an isotropic wavelet transform. Time-series changes in vessel diameter, arteriovenous ratio (AVR), and pulse waveforms were extracted from time-series changes in vessel diameter. The acquired images enabled clear visualization of retinal structures, such as the optic disc and blood vessels, and the measured AVR values ranged from 0.67 to 0.88 across five healthy participants. Furthermore, the retinal pulse waveforms showed a weak correlation with the average inter-peak intervals of the fingertip photoplethysmography signals, suggesting the system can capture cardiac-induced periodicity. These preliminary findings also suggest that the system has potential for extracting physiological information. Unlike conventional large and expensive devices using visible light, the proposed system offers a less burdensome, non-invasive alternative suitable for home-based applications. In future work, we aim to increase the number of participants and conduct more clinically oriented studies to validate the system.

In this study, we have developed a maneuvering support system for an air cushion vehicle (ACV) model. A driver can maneuver the ACV model without its drift motion by using the proposed maneuvering support system. The ACV model aims to track the trajectory which is generated from the driver’s maneuvering inputs. We refer to the trajectory as a maneuvering trajectory. In our previous study, the behavior of the ACV model became frequently unstable and it resulted in being a runaway ACV model in the case of the presence of some disturbance. In this paper, we propose an anti-runaway system for the ACV model in the case of the presence of some disturbance. When the deviation of the actual trajectory of the ACV model from the maneuvering trajectory becomes significant, the maneuvering trajectory is reset to align with the actual trajectory. Then we investigate the effects of the anti-runaway system on the maneuverability of the ACV model.

Active object search (AOS) is an approach used in mobile robots for object search. Based on the type of sensors equipped on the robot, AOS methods can be broadly classified into those primarily relying on visual information, known as active visual search (AVS), and those leveraging RF positioning data. Existing research has proposed object search systems that integrate visual and RF information. By combining these two types of data, such systems mitigate the individual limitations of each approach, thereby improving the efficiency of object search. Nonetheless, when objects of identical appearance are located nearby within the search space, they become indiscernible, posing a significant challenge to the search process’s efficacy. This paper proposes a new AOS system that combines AVS with an indoor positioning system using ultra-wideband (UWB) to address this challenge. This approach enables efficient object search and ensures accurate identification of target objects, even when visually indistinguishable objects are located near one another. Experimental results show that the system can search for the target object regardless of the initial posture and can acquire the object’s three-dimensional appearance as voxel information. Furthermore, the system successfully identified the target object even when the experimenters placed a visually indistinguishable object beside it.

In high-mix industrial environments, where task requirements change frequently, motion planners must adapt rapidly without extensive retraining or manual redesign. This study presents a state machine construction method that incorporates the previously proposed Path-Reuse (PR) algorithm. PR stores past robot trajectories and efficiently adapts them to new start–goal pairs with minimal computation, enabling the autonomous generation of safe and executable paths for articulated robots. By leveraging PR, the proposed method eliminates both the large datasets and expert intervention that are often required by learning-based approaches. Experiments in dynamically changing work cells compared the method with state machines that employ conventional planners such as RRT-connect and STOMP. The state machine incorporating the PR method significantly shortened planning times and increased success rates in collision-rich conditions.

The phenomenon of self-organization, where structures spontaneously form without external instruction, is ubiquitous in nature and holds immense potential for engineering fields. However, elucidating the self-organizing mechanisms behind the complex, hierarchical structures seen in biological organisms remains a critical challenge. This study addresses this by proposing a hierarchical extension of the Ishida model, a two-dimensional cellular automaton (CA), to generate complex forms through simple, localized interactions between neighboring elements. The model employs a multi-level hierarchical structure where transition rules are applied sequentially to each layer, often with a time lag, to promote stable pattern formation and prevent unintended side effects. By adjusting a small set of core parameters (notably w₁ and w₂), the model successfully generated a variety of static and dynamic hierarchical patterns. Specifically, it first produced a double-ring structure, which was then internally segmented into patterns with two, four, or six divisions. Furthermore, using an asymmetric initial configuration, the model generated dynamic patterns that maintained their shape while in motion, as well as forms with controlled, elongating protrusions. These results demonstrate that simple, iterative operations can create diverse, complex morphologies, offering a valuable framework for synthetic morphology and potential applications in fields like modular swarm robotics.

This study proposes a novel pre-training method employing the 3D formula-driven supervised learning for classifying three-dimensional computed tomography (CT) images, specifically for staging chronic obstructive pulmonary disease (COPD). The proposed method leverages mathematically generated 3D images as a pre-training dataset to enhance feature extraction while addressing the scarcity of labeled training data, primarily caused by ethical concerns inherent in medical imaging. This paper evaluates the effectiveness of the proposed method in two CT image classification tasks: COVID-19 classification and COPD staging. The experimental result shows that the performance of a COVID-19 classification model pre-trained using mathematically generated 3D images is competitive to the existing model pre-trained using numerous CT lung images. The result also reveals the potential of the proposed approach for COPD staging, mitigating the data scarcity issue.

Boredom is a complex psychological state, often seen as a double-edged sword. It can promote creativity, but is also linked to negative effects, such as irritability, fatigue, and even self-directed aggression. To better understand the neural underpinnings of boredom, this study applies Integrated Information Theory (IIT) to investigate how brain–body interactions reflect this state. Using electroencephalography (EEG) and autonomic measures (electrodermal activity and electrocardiography), we examined participants engaged in arithmetic tasks of varying difficulty levels. Our findings reveal that system integration, quantified by integrated information ((Phi)), increased during both easy and difficult tasks compared to the moderate condition. In contrast, the dynamical properties of the main complex—a core subset within IIT—showed an opposite trend, indicating improved body–brain coupling specifically during moderate tasks. Moreover, analyses using Earth Mover’s Distance revealed significant shifts in the main complex’s distribution during moderate tasks, whilst there were minimal changes under the easy condition. Interestingly, subjective reports indicated that irritation and tiredness were linked to transitions in the main complex, whilst boredom showed no correlation with (Phi). These results suggest that boredom may emerge from a decoupling between brain and body systems, despite heightened system integration. Conversely, stronger body–brain coupling—characteristic of the moderate task condition—may buffer against subjective stress. This research highlights the potential of IIT for quantifying internal states like boredom and emphasises the need for further studies with larger, more diverse samples to validate and generalise these findings.

In this study, we focus on the realization of a power steering control that recovers the maneuverability of the bicycle which is deteriorated due to loading baggage in the front basket, by assisting the handlebar operation. According to the analysis of the effect of loading children and baggage in the front basket on maneuverability, it is found that the cause is an increase in the moment of inertia around the handlebar axis and a change in the position of the center of gravity. To solve the issue, previous studies have proposed a power steering mechanism that applies an assist torque proportional to the user’s steering torque to the handlebar axis. This mechanism has been shown to recover reduced maneuverability. However, when a rider rides a bicycle, the rider does not only operate the handlebars but also tilts the bicycle in the roll direction. Therefore, in this study, to propose a control method to assist the steering torque of the handlebar, a mathematical model of the bicycle is considered that includes both handlebar operation and bicycle tilt. To verify the effectiveness of the proposed model, we develop an electrically power assisted bicycle with the power steering mechanism and implement an assist algorithm. The experimental results confirmed that the proposed method is effective in assisting bicyclists with luggage in the front basket of their bicycles.