Development of a Cochlear-Mimicking MEMS Sensor to Measure Auditory Stimulation in a Car

Riku Ito (Gunma University)･Kou Sasaki･Hiroyoshi Inaba (SUBARU)･Tetsuji Koyama (Gunma University)･Toshihiko Kozai (SUBARU / Gunma University)･Sinyoung Lee (University of Yamanashi)･Takuji Koike (The University of Electro-Communications)･Yuya Tanaka (Gunma University)･Tsutomu Iwase (SUBARU / Gunma University)･Takaaki Suzuki (Gunma University)

Human feeling still is not clear during the driving a car. In this reserch, we propose a cochelear-mimicking MEMS (Micro Electro Mechanical Systems) sensor to quantify the sense of reality that drivers feel with their ears to collect data for remote driving. The response of the proposed MEMS sensor was evaluated under both air and bone conduction inputs cased by automobile vibration.

Driver State Estimation from Automotive Operation Signals based on Pre-training using Vector Quantized Variational Autoencoders

Kenshiro Hanai･Yoshihiko Nankaku (Nagoya Institute of Technology)･Atsunobu Kaminuma (International Professional University of Technology in Tokyo / Nissan Motor)

This study aims to build a driver state estimator for in-vehicle services. The training of the estimator requires automotive operation signals and driver state labels, however the state labeling requires manual work. Therefore, we propose a framework in which a pre-training model is learned from a large amount of automotive operation signals and the estimator is trained using a small amount of labeled data.

Analysis, Modeling, and Control of Merging Behavior at Low Speeds

Tatsuya Ishiguro･Hiroyuki Okuda (Nagoya University)･Kenta Tominaga (Mitsubishi Electric)･Tatsuya Suzuki (Nagoya University)

In a merging environment where automatic and manual operations are mixed, it is important for both parties to communicate with each other. While much research has been conducted at high speeds, such as on expressways, little progress has been made at low speeds, such as on ordinary roads. In this presentation, we will experimentally observe the driving characteristics of drivers in right-angle merging at low speeds, and construct a right-angle merging model that takes negotiation into account.

Dynamic Estimation of Driver Model Parameters for Precise Reproduction of Driving Behavior

Ryuya Seki･Hironori Suzuki (Toyo University)

To elucidate the mechanisms of traffic accidents, it is necessary to estimate the driver's characteristics and appropriately select parameters according to the situation, ensuring an accurate reproduction of the driver's behavior. This paper aims to precisely estimate driving characteristics represented by driver model parameters based on measurements of headway distance, velocity, and relative speed to the preceding vehicle. Using a dual particle filter, we perform online estimation of the driver model parameters and convert these parameters into probability density functions. We then validate whether the parameters selected from this probability density function during simulations can accurately reproduce the driver’s behavior.

Research on Low Arousal Detection Systems with LSTM Models

Xupeng Zhou･Shuncong Shen･Toshiya Hirose (Shibaura Institute of Technology)

This study aimed to construct a Long Short-Term Memory (LSTM) model of a driver's arousal state during autonomous driving based on EEG using eye features. In the experiment, a driving simulator, eye movement meter, and electroencephalograph were used to measure eye features, in a late-night highway autonomous driving scenario. In addition, EEG data also measured band power for different frequency bands. Using the measured data, LSTM models were constructed for individual drivers, and their accuracy quantitatively evaluated by comparing the EEG output results of the constructed models with actual EEG data.