Artificial Life and Robotics最新文献

In recent years, various approaches have been proposed to design control systems that directly utilize data without mathematical plant models. Data-driven control involves updating or redesigning a controller using actual operating data, enabling fine-tuning control systems and achieving desired characteristics. However, the increasing prevalence of cyber-attacks targeting control systems presents significant societal challenges. A study by Russo and Proutiere (in Proceeding of American Control Conference (ACC), 2021) showed a poisoning approach targeting virtual reference feedback tuning, a data-driven control method. The study suggests that compromising the data used in the data-driven method may result in the closed-loop performance failing to achieve desired specifications and, in the worst case, destabilizing the control system. Hence, investigating the adverse effects of cyber-attacks on data employed in data-driven methods becomes crucial. This study explores the impact of a poisoning attack on the data used in the data-driven control method, specifically emphasizing virtual internal model tuning as a representative data-driven control approach.

Sampling-based path planning algorithms suffer from heavy reliance on uniform sampling, which accounts for unreliable and time-consuming performance, especially in complex environments. Recently, neural-network-driven methods predict regions as sampling domains to realize a non-uniform sampling and reduce calculation time. However, the accuracy of region prediction hinders further improvement. We propose a sampling-based algorithm, abbreviated to Region Prediction Neural Network RRT* (RPNN-RRT*), to rapidly obtain the optimal path based on a high-accuracy region prediction. First, we implement a region prediction neural network (RPNN), to predict accurate regions for the RPNN-RRT*. A full-layer channel-wise attention module is employed to enhance the feature fusion in the concatenation between the encoder and decoder. Moreover, a three-level hierarchy loss is designed to learn the pixel-wise, map-wise, and patch-wise features. A dataset, named Complex Environment Motion Planning, is established to test the performance in complex environments. Ablation studies and test results show that a high accuracy of 89.13% is achieved by the RPNN for region prediction, compared with other region prediction models. In addition, the RPNN-RRT* performs in different complex scenarios, demonstrating significant and reliable superiority in terms of the calculation time, sampling efficiency, and success rate for optimal path planning.

Defect detection in various industrial products ensures product quality and safety. This paper introduces an innovative design, training, and evaluation application employing CNN, CAE, YOLO, FCN, and SVM models, to facilitate defect detection without requiring extensive IT expertise. However, conventional usage of Grad-CAM for visualizing defect regions sometimes includes irrelevant areas unrelated to the target defects. A novel data augmentation technique called random masking is proposed to enhance the visualization of defective regions, leading to more accurate and focused defect detection in various industrial products. This technique is used during training, replacing non-target areas in each image with randomly generated mask patterns. The efficacy of the proposed technique is demonstrated through visualization tests of defective regions using Grad-CAM. Furthermore, an ablation study is conducted to assess the effectiveness of the data augmentation techniques, comparing the performance of Grad-CAM with and without random masking augmentation. We further provide insights into the dataset used and present noteworthy findings from the evaluation, showcasing the contributions of our work in advancing defect detection methodologies.

The authors are studying to mimic the mechanism of gait generation in animals and implement the mechanism in robots. Previously, the authors developed a quadruped robot system that spontaneously generates gait through feedback on foot pressure. The neuromorphic circuits that mimic the animal's nervous systems control the legs of the quadruped robot. Neuromorphic circuits are analog electronic circuits that output electrical spikes the same as biological neurons. However, quadruped robots system require digital processing by a microcontroller to transmit signals from pressure sensors to the neuromorphic circuit. In addition, a microcontroller drives the servo motors using a pulsewidth modulation waveform. This paper describes a newly developed neuromorphic circuit that does not require the processing of a microcontroller to convert pressure sensor signals. The authors add the receptor cell model and the integral circuit to the conventional neuromorphic circuit. By converting pressure sensor signals with the receptor cell model and the integral circuit, the neuromorphic circuit can be processed similar to a microcontroller. As a simulation and a measurement result, we confirmed that the proposed neuromorphic circuit could implement in a quadruped robot system.

Among subsets of natural language, spatial language, more exactly spatiotemporal language here, has been considered most essential for human-like interaction between people and robots expected in near future. Quite distinctively from conventional learning-based approaches to natural language understanding (NLU), mental-image-directed theory (MIDST) proposes a robotic deep NLU methodology based on a mental image model as a formal system for knowledge representation and reasoning. The application system named CoMaS is designed to understand User’s utterances in text and respond in text or animation through human-like spatiotemporal reasoning based on the mental image model. In this work, CoMaS was compared with human subjects through a psychological experiment on spatiotemporal language understanding and showed globally good agreement with them and locally some interesting and reasonable disagreement. This kind of disagreement was found among the human participants as well and explainable as difference in personal conceptualization or reasoning based on mental image. The experimental results and theoretical discussion based on them showed well the effectiveness and uniqueness of our study.

This paper proposes a method to enables the generation of short-length routes with consideration of obstacle avoidance and significantly reduces the computation time compared to existing research for ocean route optimization. The reduced computation time allows recalculation of routes for autonomous vessel underway. By simulating the recalculation of four cases of the vessel underway that may require recalculation, this paper demonstrates that the proposed method can generate new and superior routes for the vessel that needs to change their routes due to certain factors.

In this paper, we propose a control strategy for non-prehensile object transport using a multi-robot system. While an object can be unmanageable for a single robot to push and transport, we demonstrate via simulations that a team of cooperative robots can be used to transport such an object. The proposed control strategy is divided into two phases: caging and cooperative transport. In the first phase, the robots start from arbitrary positions and then approach the object to be transported, forming a cage around it. The second phase consists of cooperatively transporting the object ensuring it remains caged during transport. In the proposed strategy, the robots take a decentralized approach where robots behave autonomously while being in indirect communication by leveraging distributed data structures to share their state. Our use of distributed data structures like distributed locks, sets, and maps offered by the data grid concept provides a mechanism for inter-robot communication without development of a new application-specific protocol. To our knowledge, the use of in-memory data grid (IMDG) is new to the field of multi-robot systems. We believe it could provide a promising solution to simplify inter-robot communication. In this paper, we present our design for a coordinated motion control strategy for object transport leveraging IMDG. Finally, we demonstrate our results using a realistic simulator that shows the feasibility of our approach in various environments.

This paper demonstrates a controller design of a multi-legged robotic swarm in a rough terrain environment. Many studies in swarm robotics are conducted with mobile robots that work in relatively flat fields. This paper focuses on a multi-legged robotic swarm, which is expected to operate not only in a flat field but also in rough terrain environments. However, designing a robot controller becomes a challenging problem because a designer has to consider how to coordinate a large number of joints in a robot, besides the complexity of a swarm problem. This paper employed an evolutionary robotics approach for the automatic design of a robot controller. The experiments were conducted by computer simulations with the path formation task. The results showed that the proposed approach succeeds in generating collective behavior in flat and rough terrain environments.

A multi-agent system is a collection of autonomous, interacting agents that share a common environment. These entities observe their environment using sensors and interact with the environment. A multi-agent system that develops cooperative strategies by reinforcement learning does not perform well, mostly because of the sparse reward problem. This study conducts a 3D environment in which robots play the beach volleyball game. This study combines imitation learning (IL) with reinforcement learning (RL) to solve the sparse reward problem. The results show that the proposed approach gets a higher score in the Elo rating system and robots perform better than the conventional RL approach.

There is an increasing need for methods to estimate autonomic nerve activity, such as using autonomic nerve activity as an indicator of stress and to improve the activity environment, such that people can live more comfortably. In a previous study, a method for estimating autonomic nervous activity using facial thermal images was established. Such activity was evaluated by obtaining the differential temperature of the nasal region, which is a sympathetic index of autonomic nervous activity, and the less-affected forehead region. However, this method requires the use of an expensive far-infrared camera, which is difficult to obtain and is cited as a problem. Another study suggested acquiring pulse waves and heartbeats by obtaining changes in blood flow from the heart using real facial images. This was obtained using independent component analysis for each Red–Green–Blue (RGB [color model]) component value of the real image. However, the problem with this method is that it cannot evaluate autonomic nerve activity, such as thermal images, because it cannot examine changes in the blood flow in peripheral blood vessels. Therefore, in this study, a method for estimating autonomic nerve activity is proposed, with the same accuracy as thermal images, by obtaining changes in peripheral blood flow from real images taken with an easily usable web camera. To obtain the changes in peripheral blood flow from real images, we used the property that the incident depth of light entering the skin differs depending on the color component. By considering the difference between the R component (which enters the skin at the deepest depth) and the B component (which is almost reflected by the epidermis), we assumed that we could capture the blood flow in the widest range and measure the change in blood flow in the peripheral vasculature. To cause unpleasant irritation, an experiment was conducted in which participants performed a randomly assigned memorization task for 10 min. A rest period of 1 min was provided before and after the mental calculation task, and a subjective evaluation questionnaire was administered during the rest period and before the mental calculation task. The results showed that some subjects exhibited a significant decrease in the R-B component values when they noticed calculation errors, whereas others exhibited an increase in the R-B component values after completing the memorization task. Also, some subjects showed a correspondence between the variation of R-B component values and the results of the questionnaire. Additionally, compared to the conventional method, it was clear that we could obtain real-time and even fine variations in response to event switching. These results suggest that it is possible to evaluate autonomic nervous system activity from real image data to the same extent as from thermal images.