Panasonic Corporation

Considering teleoperation of manipulators, we examined the effects of latency that may occur during teleoperation on the sense of agency. The study assumed pick-and-place operations with the manipulator, and initially evaluated the impact of steady latency on the task. Next, in order to further investigate the effect of latency in which area in the operation, we divided the pick-and-place operation into a rough operation, which is a part of large movement, and a precision operation, which is a fine positioning operation necessary when grasping an object, and evaluated the difference in the effect of latency in each operation. In addition to the steady-state latency, the evaluation also took into account the latency fluctuations that may occur in actual communication. In addition, we analyzed the changes in subjective evaluation due to delay and the operation logs, clarified the factor structure of the sense of subjectivity, and also showed the relationship with the logs.

The magnetic field detection based on the interference phenomenon of surface-mode spin waves has been demonstrated in yttrium iron garnet (YIG) thin films, where the asymmetric arrangement of two excitation sources and one detection antenna allows for the field detection with a simple YIG strip structure that does not require microfabrication. The magnetic field can be detected by observing changes in the amplitude of the standing wave at the detection position, which result from alterations in the wavenumber of the excited spin wave caused by variations in the magnetic field. Time-domain measurements confirmed that the interference signal of the spin wave changed with the magnetic field. The induced electromotive force yielded a change of approximately 7 mV for a magnetic field change of ± 0.13 mT, resulting in a sensitivity of 24–25 V/T. The sinusoidal interference calculation using the wavenumber change due to a small magnetic field derived from the dispersion relation of spin waves agrees with the experimental results. This suggests that the mechanism of magnetic field detection is the wavenumber change due to the magnetic field.

Background: Falls are among the most prevalent workplace accidents, necessitating thorough screening for susceptibility to falls and customization of individualized fall prevention programs. The aim of this study was to develop and validate a high fall risk prediction model using machine learning (ML) and video-based first three steps in middle-aged workers. Methods: Train data (n=190, age 54.5±7.7 years, 48.9% male) and validation data (n=28, age 52.3±6.0 years, 53.6% male) were enrolled in this study. Pose estimation was performed using a marker-free deep pose estimation method called MediaPipe Pose. The first three steps, including the movements of the arms, legs, trunk, and pelvis, were recorded using an RGB camera, and the gait features were identified. Using these gait features and fall histories, a stratified K-fold was used to ensure balanced training and test data, and the area under the curve (AUC) and 95% confidence interval (CI) were calculated. Results: Of 77 gait features in the first three steps, we found 3 gait features in men with an AUC of 0.909 (95% CI, 0.879-0.939) for fall risk, indicating an 'Excellent' (0.9-1.0) classification, while we determined 5 gait features in women with an AUC of 0.670 (95% CI, 0.621-0.719), indicating a 'sufficient' (0.6-0.7) classification. Conclusions: These findings suggest that fall risk prediction can be developed based on ML and the first three steps in men; however, the accuracy was only sufficient in men. Further development of the formula is required for women to improve its accuracy in the middle-aged working population.

In the beyond 5G/6G, the usage of the wideband transmission in the higher frequency band is expected. In this paper, we propose a method to compensate IQ imbalance caused by the path length differences between I and Q channels in the direct conversion transmitter in the wideband OFDM systems. In the proposed method, compensation signals are generated using the path length difference and the modulation symbols transmitted on the paired subcarriers which are known in the transmitter side. We analyzed the IQ imbalance and the proposed compensation method, then the effects of the proposed method are explained through the simulation.

For humans and robots to coexist and cooperate, robots need to be inexpensive to lower the hurdle for introduction, and they need to make gentle contact with humans, objects, and the environment, so force control is a key technology. Torque sensor feedback is used to achieve contact, but the mounted torque sensor must be able to detect both small forces to realize skillful work while contacting an object and large forces such as collision. In this research, we focus on a method of using thrust displacement in the worm reduction mechanism when load torque is applied to the worm gear for sensing, and develop a low-cost, space-saving, and lightweight joint module in which the torque sensing function and drive unit are integrated without the need for additional sensor devices. Furthermore, two types of elastic support structures are used to provide two-stage characteristics to achieve fine to wide range torque sensing.

We propose a pair of step-climbing units for manual wheelchair users who have no disabled upper limbs and hopes independent life without caregivers. The unit is a rectangle shape, which comprises a DC motor and a crawler belt. These devices are used combining with a pair of mobile ramps. Each unit moves on each ramp producing propelling force to rear wheels of a wheelchair from both sides. Some of wheelchair users go to work by car. It is important to load the units and the ramps into the car easily. We also make a pair of light and compact prototype ramps using GFRP and aluminum pipes. We discuss the control of the units through wireless communication when descending. Outdoor experimental results when climbing up and down two steps, which height was 19 cm each, confirmed effectiveness of the design.

To achieve disassembly planning for large home appliances, we propose to simultaneously plan not only two-finger grasp, suction grasp, and contact by multiple end-effectors, but also the trajectory of a robot arm and rotational motion of an automatic rotation stage. This study assumes the combined use of multiple hardware devices to manipulate various shapes for the disassembly of large home appliances. To reduce the search range in the simultaneous planning, we propose the use of intermediate points for the trajectory and prioritized exploration of rotation angles of the rotation stage according to the disassembly task. Our experimental results demonstrated the effectiveness of the proposed methods by validating the generated plans in simulations and their executions in the real world.

The generalization performance of machine learning models deteriorates when the models are trained with mislabeled data. Existing methods to address mislabeled data rely on pre-processing or in-processing of the training. However, those methods require retraining when applied to trained models. As the model size and dataset size increase, the cost of retraining the model becomes a significant issue, necessitating the development of new approaches. In this paper, we propose a new method to remove mislabeled data from trained models without retraining via machine unlearning. Our proposed method consists of two stages: first, detecting mislabeled data from trained models, and second, unlearning these data from the models. We conduct extensive experiments on the MNIST dataset to evaluate our proposed method. To comprehensively evaluate the effectiveness of our proposed method, we perform individual experiments for the detection stage and the unlearning stage. Our findings demonstrate that the detection stage performs well when the proportion of mislabeled data is low, and the unlearning stage effectively enhances model accuracy. However, in an integrated experiment involving both stages,we observed intriguing yet negative results: despite the effectiveness of individual stages, model accuracy did not improve due to the high proportion of mislabeled data. Our code is available at https://github.com/speed1313/mislabel-unlearning.