In this study, we propose a reconfigurable laminate mechanism based bistable tail for Throwbot transforming into a ball type and a wheel type. Various robots such as snake robots, drones, and throwing robots for life-saving missions on behalf of humans at disaster sites have been studied. In particular the hybrid type throwing robot can have both the throwing ease of the ball type and the driving stability of the wheel type. However, it requires the tail to be stored inside when being thrown and to be rigidly deployed when driving. To satisfy these requirements, we developed a foldable tail based on scissor lift structure in our previous study. But, such a structure was composed of only rigid parts, which caused interference with other parts when stored, and difficulty about changing the maximum deployed tail length further. To overcome these limitations, we wanted to develop a bistable tail suitable for the hybrid type that can maintain a bendable state and a rigid state. Before actual development, we calculate the minimum tail length for overcoming obstacle through statics analysis. Then, we design a bistable structure utilizing a reconfigurable laminate mechanism. Next, we calculate the design constraints to mount it on the actual robot. Finally, the developed tail is mounted on the actual Throwbot to perform obstacle overcoming experiments. We confirm that it can secure both ease throwing and stable obstacle overcoming ability. Through this, we propose a bistable tail suitable for the hybrid type throwing robots.

In the field of predicting the remaining useful lifetime (RUL) of equipment based on health indicators (HIs), the effective extraction of equipment status poses a persistent challenge. Numerous studies have focused on the extraction of HI of an equipment, frequently proposing training methods that utilize one-dimensional latent space autoencoders for computing the loss function with HI. This was imperative due to the compositional nature of HIs are real numbers. However, in cases where equipment status exhibits nonlinear and intricate structuring, the latent vector necessitates representation within a sufficiently expansive space. This paper introduces a methodology for mapping real numbers to higher dimensions to achieve a more efficient representation of HI. Upon applying our proposed methodology to various HI extraction models, we observed that in most instances, the performance of HI extraction was enhanced. Ultimately, this contributed to an improvement in the prediction accuracy to RUL. The efficacy of our learning mechanism was validated across four subsets (FD001 to FD004) of the C-MAPSS dataset provided by NASA. Notably, the methodology presented in this study holds significance as a learning mechanism adaptable to various approaches.

This paper proposes a point cloud (PC) visual quality assessment (VQA) framework that reflects the human visual system (HVS). The proposed framework compares natural images acquired using a digital camera and PC images generated via 2D projection in terms of appropriate objective quality evaluation metrics. Humans primarily consume natural images; thus, human knowledge is typically formed from natural images. Thus, natural images can be more reliable reference data than PC data. The proposed framework performs an image alignment process based on feature matching and image warping to use the natural images as a reference which enhances the similarities of the acquired natural and corresponding PC images. The framework facilitates identifying which objective VQA metrics can be used to reflect the HVS effectively. We constructed a database of natural images and three PC image qualities, and objective and subjective VQAs were conducted. The experimental result demonstrates that the acceptable consistency among different PC qualities appears in the metrics that compare the global structural similarity of images. We found that the SSIM, MAD, and GMSD achieved remarkable Spearman rank-order correlation coefficient scores of 0.882, 0.871, and 0.930, respectively. Thus, the proposed framework can reflect the HVS by comparing the global structural similarity between PC and natural reference images.

To understand and mimic the working mechanism of the brain, neuroscientists rely on brain simulations that operate in a time-driven manner. The simulation involves evaluating how the neurons change their states over time and transferring spikes to the connected neurons through synapses. It also simulates learning by evaluating how the synapses change their weights according to the spiking activity of the neurons. To explore various behaviors of the brain and thus make great advances, neuroscientists need a methodology to support large-scale simulations in both an accurate and efficient manner. For accurate simulations, existing simulators adopt a time-precise simulation methodology where the simulator computes all the neuronal and the synaptic state changes in time order. Unfortunately, they suffer from significant underutilization and energy inefficiency as the simulator scales.In this paper, we present NeuroSync, a fast, energy-efficient, and scalable hardware-based accelerator for accurate brain simulations. The key idea is to adopt a speculative simulation methodology at a minimum overhead along with architectural support. NeuroSync achieves high efficiency using an optimal dataflow for the speculative simulations. At the same time, it ensures simulation accuracy by carefully designing a rollback and recovery mechanism to handle mis-speculations. To implement the methodology at a low cost, NeuroSync further proposes a speculation-optimal learning simulation method. Our evaluations show that 64-chip NeuroSync achieves 3.37× speedup and 3.81× higher energy efficiency with only 10.96% area overhead. The evaluations also show that NeuroSync is extremely scalable with higher speedup as the system scales.

The crossbar structure of the nonvolatile memory enables highly parallel and energy-efficient analog matrix-vector-multiply (MVM) operations. To exploit its efficiency, existing works design a mixed-signal deep neural network (DNN) accelerator, which offloads low-precision MVM operations to the memory array. However, they fail to accurately and efficiently support the low-precision networks due to their naive ADC designs. In addition, they cannot be applied to the latest technology nodes due to their premature RRAM-based memory array.In this work, we present 3D-FPIM, an energy-efficient and robust mixed-signal DNN acceleration system. 3D-FPIM is a full-stack 3D NAND flash-based architecture to accurately deploy low-precision networks. We design the hardware stack by carefully architecting a specialized analog-to-digital conversion method and utilizing the three-dimensional structure to achieve high accuracy, energy efficiency, and robustness. To accurately and efficiently deploy the networks, we provide a DNN retraining framework and a customized compiler. For evaluation, we implement an industry-validated circuit-level simulator. The result shows that 3D-FPIM achieves an average of 2.09x higher performance per area and 13.18x higher energy efficiency compared to the baseline 2D RRAM-based accelerator.

Recently, the demand for high-quality video content has rapidly been increasing, led by the development of network technology and the growth in video streaming platforms. In particular, displays with a high refresh rate, such as 120 Hz, have become popular. However, the visual quality is only enhanced if the video stream is produced at the same high frame rate. For the high quality, conventional videos with a low frame rate should be converted into a high frame rate in real time. This paper introduces a bidirectional intermediate flow estimation method for real-time video frame interpolation. A bidirectional intermediate optical flow is directly estimated to predict an accurate intermediate frame. For real-time processing, multiple frames are interpolated with a single intermediate optical flow and parts of the network are implemented in 16-bit floating-point precision. Perceptual loss is also applied to improve the cognitive performance of the interpolated frames. The experimental results showed a high prediction accuracy of 35.54 dB on the Vimeo90K triplet benchmark dataset. The interpolation speed of 84 fps was achieved for 480p resolution.