Artificial Life and Robotics最新文献

Robots have been widely used in daily life in recent years. Unlike conventional robots made of rigid materials, soft robots utilize stretchable and flexible materials, allowing flexible movements similar to those of living organisms, which are difficult for traditional robots. Previous studies have used periodic signals to control soft robots, which lead to repetitive motions and make it challenging to generate environment-adapted motions. To address this issue, control methods can be learned through deep reinforcement learning to enable soft robots to select appropriate actions based on observations, improving their adaptability to environmental changes. In addition, as mobile robots have limited onboard resources, it is necessary to conserve battery consumption and achieve low-power control. Therefore, the use of spiking neural networks (SNNs) with neuromorphic chips enables low-power control of soft robots. In this study, we investigated the learning methods for SNNs aimed at controlling soft robots. Experiments were conducted using a caterpillar-like soft robot model based on previous studies, and the effectiveness of the learning method was evaluated.

This paper proposes a drowsiness expression rating system that can rate drowsiness in real time using only video information. Drowsiness in drivers is caused by various factors, including driving on monotonous roads, and can lead to numerous problems, e.g., traffic accidents. Previously, we developed an offline drowsiness evaluation system the uses only video image information from MediaPipe, which is a marker-less facial motion capture system. The proposed system can perform real-time drowsiness rating on multiple platforms and requires a smartphone or personal computer. Results of applied to car driving demonstrate that the accuracy of the proposed system was 89.7%, 78.8%, and 65.0% for binary, three-class, and five-class classification tasks, respectively. In addition, the proposed system outperformed existing systems in binary, three-class, and five-class classification tasks by 6.0%, 0.8%, and 4.3%, respectively. These results demonstrate that the proposed system exhibits a higher accuracy rate than the existing methods.

This study proposes a method for determining the appropriate harvesting order for densely growing wood ear mushrooms by recognizing their growth stages and harvesting priorities from depth images obtained from a stereo camera. We aim to minimize crop damage and improve the quality of harvested crops during the harvesting of densely growing crops using a robot arm. The proposed two-stage method consists of two models—one of the models to recognize priority harvest regions, and the other model to identify individual wood ear mushroom regions and growth stages. The final harvesting order is determined based on the outputs of these models. The models were trained using simulated CGI data of wood ear mushroom growth. The experimental results show that the appropriate harvesting order can be outputted in 57.5% of the cases for the 40 sets of test data. The results show that it is possible to determine the harvesting order of dense wood ear mushrooms based solely on depth images. However, there is still room for improvement in operations in actual environments. Further work is needed to enhance the method’s robustness and accuracy.

Every year, cardiovascular diseases cause millions of deaths worldwide. These diseases involve complex mechanisms that are difficult to study. To remedy this problem, we propose to develop a heart–brain platform capable of reproducing the mechanisms involved in generating the heartbeat. The platform will be designed to operate in real time, with the most economical and integrated design possible. To achieve this, we are implementing highly biologically coherent cellular models on FPGA, which we interconnect with in vitro cell cultures. In our case, we are using the Maltsev–Lakatta cell model, which describes the behavior of the pacemaker cells responsible for the heart rhythm, to stimulate a cardiomyocyte culture.

Neurological disorders affect millions globally and necessitate advanced treatments, especially with an aging population. Brain Machine Interfaces (BMIs) and neuroprostheses show promise in addressing disabilities by mimicking biological dynamics through biomimetic Spiking Neural Networks (SNNs). Central Pattern Generators (CPGs) are small neural networks that, emulated through biomimetic networks, can replicate specific locomotion patterns. Our proposal involves a real-time implementation of a biomimetic SNN on FPGA, utilizing biomimetic models for neurons, synaptic receptors and synaptic plasticity. The system, integrated into a snake-like mobile robot where the neuronal activity is responsible for its locomotion, offers a versatile platform to study spinal cord injuries. Lastly, we present a preliminary closed-loop experiment involving bidirectional interaction between the artificial neural network and biological neuronal cells, paving the way for bio-hybrid robots and insights into neural population functioning.

We have attempted to estimate blood glucose levels based on facial images measured in the near-infrared band, which is highly biopermeable, to establish a remote minimally invasive blood glucose measurement method. We measured facial images in the near-infrared wavelength range of 760–1650 nm, and constructed a general model for blood glucose level estimation by linear regression using the weights of spatial features of the measured facial images as explanatory variables. The results showed that the accuracy values of blood glucose estimation in the generalization performance evaluation were 43.02 mg/dL for NIR-I (760–1100 nm) and 43.61 mg/dL for NIR-II (1050–1650 nm) in the RMSE of the general model. Since biological information is nonlinear, it is necessary to explore suitable modeling methods for blood glucose estimation, including not only linear regression but also nonlinear regression. The purpose of this study is to explore suitable regression methods among linear and nonlinear regression methods to construct a blood glucose estimation model based on facial images with wavelengths from 760 to 1650 nm. The results showed that model using Random Forest had the best estimation accuracy with an RMSE of 36.02 mg/dL in NIR-I and the MR model had the best estimation accuracy with RMSE of 36.70 mg/dL in NIR-II under the current number of subjects and measurement data points. The independent components selected for the model have spatial features considered to be simply individual differences that are not related to blood glucose variation.

This paper describes a method for integrating multiple dense point clouds using a shared landmark to generate a single real-scale integrated result for photogrammetry. It is difficult to integrate high-density point clouds reconstructed by photogrammetry because the scale differs with each photogrammetry. To solve this problem, this study places a QR code of known sizes, which is a shared landmark, in the reconstruction target environment and divides the reconstruction target environment based on the position of the QR code that is placed. Then, photogrammetry is performed for each divided environment to obtain each high-density point cloud. Finally, we propose a method of scaling each high-density point cloud based on the size of the QR code and aligning each high-density point cloud as a single high-point cloud by partial-to-partial registration. To verify the effectiveness of the method, this paper compares the results obtained by applying all images to photogrammetry with those obtained by the proposed method in terms of accuracy and computation time. In this verification, ideal images generated by simulation and images obtained in real environments are applied to photogrammetry. We clarify the relationship between the number of divided environments, the accuracy of the reconstruction result, and the computation time required for the reconstruction.

In this study, we propose a method for detecting feature points of obstacles by real-time processing using a monocular camera and a laser. Specifically, we propose an algorithm that detects laser beams emitted on obstacles by binarization, transforms pixel coordinates into distance using triangulation by a laser beam that is obliquely incident on the ground, and estimates the placement angle of the obstacle in real-time processing. No method has been devised to detect the placement angle of obstacles using a monocular camera and a laser. Furthermore, the proposed method does not calculate the linear distance from the camera, but from the position where the camera is moved horizontally parallel to the obstacle in front of the camera, thus detecting the placement angle independently of the placement position of the obstacle. As a result, it was confirmed that the placement angles of obstacles could be detected with a maximum error of (6^{circ }) in six placement positions. Therefore, we succeeded in automatically detecting the placement angle of obstacles in real-time, independent of the illumination of the measurement environment, the reflectance of the obstacle.

In English–Chinese machine translation shift, the processing of out-of-vocabulary (OOV) words has a great impact on translation quality. Aiming at OOV, this paper proposed a method based on word vector similarity, calculated the word vector similarity based on the Skip-gram model, used the most similar words to replace OOV in the source sentences, and used the replaced corpus to train the Transformer model. It was found that when the original corpus was used for training, the bilingual evaluation understudy-4 (BLEU-4) of the Transformer model on NIST2006 and NIST2008 was 37.29 and 30.73, respectively. However, when the word vector similarity was used for processing and low-frequency OOV words were retained, the BLEU-4 of the Transformer model on NIST2006 and NIST2008 was improved to 37.36 and 30.78 respectively, showing an increase. Moreover, the translation quality obtained by retaining low-frequency OOV words was better than that obtained by removing low-frequency OOV words. The experimental results prove that the English–Chinese machine translation shift method based on word vector similarity is reliable and can be applied in practice.

With the implementation of intelligent unmanned aerial vehicles (UAVs) in power equipment inspection, managing the obtained inspection results through information technology is increasingly crucial. This paper collected insulator images, including images of standard and self-exploding insulators, during the inspection process using intelligent UAVs. Then, an optimized you only look once version 5 (YOLOv5) model was developed by incorporating the convolutional block attention module and utilizing the efficient intersection-over-union loss function. The detection performance of the designed algorithm was analyzed. It was found that among different models, the YOLOv5s model exhibited the smallest size and the highest detection speed. Moreover, the optimized YOLOv5 model showed a significant improvement in speed and accuracy for insulator detection, surpassing other methods with a mean average precision of 93.81% and 145.64 frames per second. These results demonstrate the reliability of the improved YOLOv5 model and its practical applicability.