Artificial Life and Robotics最新文献

Existing electroencephalography (EEG) studies predominantly involve participants in stationary positions, which presents challenges in accurately capturing EEG data during physical activities due to motion-induced noise and artifacts. This study aims to assess and validate the efficacy of the Soft Dynamic Time Warping (Soft-DTW) clustering method for analyzing EEG data collected during physical activity, focusing on an oddball auditory task performed while walking. Employing a mobile active bio-amplifier, the study captures brain activity and assesses auditory event-related potentials (ERPs) under dynamic conditions. The comparative performance of five clustering techniques, k-shape, kernels, k-means, Dynamic Time Warping, and Soft-DTW, in terms of their effectiveness in artifact reduction, was analyzed. Results indicated a significant difference between target and non-target auditory stimuli, with the target stimuli exhibiting a positive (positive) potential, although of smaller magnitude. This outcome suggests that, despite significant artifact interference from walking, Soft-DTW facilitates extracting differences in cognitive processes for the oddball task from the EEG data.

Artificial neural networks, which mimic the neural networks of living organisms, are being applied as advanced information processing systems in various fields such as robotics. Conventional artificial neural networks use CPUs and software programs, but huge numerical computations are required to imitate a large-scale neural network. On the other hand, hardware artificial neural networks have been proposed. Hardware models neurons and synapses using analog electronic circuits, and thus can mimic the neural signals generated by neural networks without the need for numerical calculations. We have been developing a hardware artificial neural network mimicking the neural network in the human brainstem and spinal cord that is involved in gait control, and applying it to a musculoskeletal humanoid robot that mimics the human musculature and skeletal structure. In this paper, we propose an artificial spinal cord circuit for gait control of a musculoskeletal humanoid robot. Focusing on the movement of stepping over an obstacle, we confirmed through circuit simulations that the artificial spinal cord circuit can generate stepping-over patterns arbitrarily while walking and running.

With the development of virtual reality technology, the use of avatars is attracting increasing attention. Recently, the effects of various avatars in immersive virtual reality environments on users' psychological senses and behavior, such as the sense of body ownership, sense of agency, the Proteus effect, etc., have been reported and actively studied. However, the effects of using various avatars in a non-immersive environment on users' psychological senses and behavior have not yet been fully examined. In this study, we examined how avatar shapes affect the user's psychological senses and physical motions in a non-immersive environment using a penguin avatar and a smoke avatar, with each avatar having a different shape and degrees of freedom and comparing them to a human avatar. Specifically, experiments in which whole-body physical motions were performed were conducted using these three avatars, subjective psychological senses were evaluated through questionnaires, and an objective evaluation was conducted through body-tracking data. The results suggested that the avatar shapes have an effect such that the user's body motion changes unconsciously in a non-immersive environment.

To mitigate microplastic and suspended solid debris problems, various underwater debris-collecting devices have been proposed; however, due to concerns regarding blockage in these devices’ suction pumps, simple-structured pumps with high robustness are more suitable for long-term operation. Thus, we previously proposed a debris-capturing pump mechanism inspired by the jellyfish of the Rhizostomeae order’s simple anatomy, focusing on the flow around the oral arms, which is expected to greatly affect debris-collecting performance. In the current study, the vertically integrated two-dimensional jellyfish-inspired pump’s bell material and the installment angle of the rectifier plates mimicking the oral arms were varied across four configurations, and the flow fields generated by the pump with their governing dominant flow structures were investigated using particle image velocimetry (PIV) and proper orthogonal decomposition (POD) to evaluate the effect of both variables on the flow structure. Experimental results suggest that both variables affect the flow structure and reverse flow rate significantly. By increasing the bell’s elastic modulus and installing the plates at a moderate angle, the reverse flow in the bell-opening motion can be suppressed.

Using many agents with different characteristics is more effective than using a homogeneous agent to observe a large environment persistently. This study focuses on the heterogeneity of agents’ observation capabilities, such as sensor resolution, by representing these differences through probabilistic observation. This representation allows agents to compute mutual information when selecting surveillance areas and move to where they can obtain the most information from their observations. In addition, we introduce confidence decay for three or more states, a strategy to encourage agents to revisit locations that have not been observed for an extended period of time. Confidence decay represents a gradual decrease in the estimates’ reliability since the state may have changed during the unobserved period. This strategy increases the mutual information of locations that have not been observed for a long time so that the agents will move toward them. Simulations in a changing environment show that the proposed method enables heterogeneous multi-agents to perform persistent surveillance according to their observation capabilities. It also outperforms the existing partition and sweep method in a quantitative comparison of observation accuracy.

The real-time monitoring of crowd size is essential for accurate and efficient evacuation guidance and other disaster response efforts in large-scale events. Hence, we developed a portable and cost-effective crowd monitoring system with environmentally friendly features, including waterproofing and dustproofing, using Wi-Fi technology. This system can cope with media access control (MAC) address randomization in detected Wi-Fi devices to enhance headcount detection accuracy. To assess the precision of this method in crowd size estimation, we conducted comparative experiments at the large-scale event “Gorokuichi” in 2021 and 2022. The mean absolute percentage error was 5.86% in 2021 and 8.56% in 2022, demonstrating high consistency, with correlations exceeding 80% between the estimated numbers and human observer counts (true values), thus confirming the effectiveness of our system.

The distributed static synchronous series compensator can optimize the transmission capacity of the power grid. However, the research on the coordinated control and interaction between the devices is not mature enough, and it still needs to be further explored. Therefore, a coordinated control strategy based on multi-objective immune optimization algorithm is proposed in this paper. To realize the feasibility of the coordination strategy, simulation experiments were carried out. The results showed that through the coordination of multi-objective optimization artificial immune algorithm, the optimization rate of active power and reactive power of the line reached 89.88%, and the optimization rate of direct current capacitance and voltage also reached 51.45%, which confirmed the effectiveness of the coordination strategy. It can improve the application of distributed static synchronous series compensator in power grid transmission.

This paper outlines a project aimed at realizing a “Smarter Inclusive Society” by 2050 through the integration of AI robots into various public facilities. Led by the Cabinet Office’s “Moonshot Research and Development Program,” the project focuses on developing Adaptable AI-enabled Robots that enhance self-efficacy by supporting users’ abilities while maintaining their sense of independence. Key to the project is the Robotic Nimbus, a soft and flexible robot designed to provide tailored assistance while preserving user agency. The concept of “Adaptable AI-enabled Robots” is introduced to ensure versatility in accommodating user needs and preferences. In addition to physical assistance, the project emphasizes creating engaging experiences through activities like dance and sports, fostering excitement and inclusivity. Collaborations, such as the “Yes We Dance!” performance, demonstrate the potential of AI technology in enhancing rehabilitation opportunities and promoting social participation. By 2050, the project aims to establish a society where AI robots contribute to mental, physical, and social wellbeing, empowering individuals to engage in independent activities and fostering a vibrant, inclusive community. This paper is a compilation of articles/papers/presentations previously presented on the Moonshot Hirata project.

To enable successful automated driving, precise behavior prediction of surrounding vehicles is indispensable in urban traffic scenarios. Furthermore, given that a vehicle’s behavior is influenced by the movements of other road users, it becomes crucial to estimate their intentions to anticipate precise future motion. However, the elevated complexity resulting from interdependencies among traffic participants and the uncertainty arising from the object recognition errors present additional challenges. Despite extensive research on inferring intentions, many studies have concentrated on estimating intentions from interactions, resulting in a lack of practicality in urban traffic environments due to low computational efficiency and low robustness against recognition failure of strongly interacting road users. In this paper, we introduce a practical stochastic model for intention estimation and trajectory prediction of surrounding vehicles in automated driving under urban traffic environments. The trajectory is forecasted based on hierarchically computed and probabilistically estimated intentions, which represent an interpretation of vehicle behavior, utilizing only the kinematic state of the focal vehicle and HD maps to ensure real-time performance and enhance robustness. The evaluated results demonstrate that the proposed model surpasses straightforward methods in terms of accuracy while maintaining computational efficiency and exhibits robustness against the recognition failure of traffic participants which strongly influence the focal vehicle.

In communication, emotional information is crucial, yet its preservation in tweet embeddings remains a challenge. This study aims to address this gap by exploring three distinct methods for generating embedding vectors of tweets: word2vec models, pre-trained BERT models, and fine-tuned BERT models. We conducted an analysis to assess the degree to which emotional information is conserved in the resulting embedding vectors. Our findings indicate that the fine-tuned BERT model exhibits a higher level of preservation of emotional information compared to other methods. These results underscore the importance of utilizing advanced natural language processing techniques for preserving emotional context in text data, with potential implications for enhancing sentiment analysis and understanding human communication in social media contexts.