Construction of an Accelerator Operation Prediction Model Based on Driving Data Clustering Considering Driving Characteristics

Takafumi Hosogi･Ryu Murotani･Susumu Sato (Institute of Science Tokyo)

In this study, real-world driving data were clustered based on driving characteristics, and an LSTM-based accelerator operation prediction model was constructed for each cluster. The proposed method aims to perform prediction that reflects the operational tendencies of each driving dataset. The experimental results showed that cluster-specific modeling improved the prediction performance of accelerator operation.

Study on an In-Cabin Biosensing Method Using Endocrine Indicators

Masaaki Nishi (Honda R&D)･Shintaro Izumi･Nobuhito Taniguchi･Yuto Noda (University of Kobe)･Toshihide Shiino･Yasushi Noguchi (Honda R&D)

To estimate occupants’ stress and comfort/discomfort states in a vehicle cabin, this study focused on the relationship between heart rate variability (HRV) and endocrine responses. HRV analysis windows were shifted backward from each saliva sampling time to synchronize continuous HRV features with delayed salivary hormone changes. A convolutional neural network (CNN) -based estimation model was then used to predict hormone dynamics, enabling successful classification of mental and physical stress states.

Impact of Trust Formation in Driver Assistance Systems on Drivers' Peace of Mind

Yousuke Furuya･Yoshihisa Okamoto･Taisei Kamio･Hiromasa Kenmotsu･Nanae Michida (Mazda)･Norihiro Sadato (Ritsumeikan University)

To enable drivers to enjoy driving with peace of mind, it is considered essential that they have trust in vehicle safety and driver assistance functions. We hypothesized that a key requirement for building such trust is that the assistance meets drivers’ expectations. Using a driving simulator experiment, we demonstrated that the formation of trust leads to increased peace of mind during driving.

Driver Accelerator Prediction using Real-World Environmental Parameters: Integrating a Brake Prediction Mechanism

Ryu Murotani･Takafumi Hosogi･Susumu Sato (Institute of Science Tokyo)

This research constructs a driver accelerator-prediction model from real-world environmental parameters (inter-vehicle distance, relative speed, road gradient, etc.) for vehicle simulations in the development phase. The model has two parallel prediction branches based on machine learning (LSTM/Transformer) : one predicts the accelerator pedal opening, and the other predicts the probability of braking. By integrating these outputs via a gating mechanism, the model improves overall accuracy, achieving particularly high prediction performance around vehicle stops.

Effects of Airflow Stimulation Considering Body-Part Sensitivity Characteristics on Drivers' Thermal Comfort and Arousal Maintenance

Jongseong Gwak (Takushoku University)･Yoko Hayashi (Institute of Science Tokyo)･Akinari Hirao (Shibaura Institute of Technology)･Motoki Shino (Institute of Science Tokyo)

To maintain drivers’ arousal while suppressing thermal discomfort, this study focused on body-part sensitivity characteristics to airflow stimulation under thermal environments and investigated the effects of airflow stimulation to the neck on thermal comfort and arousal level. The results indicated that, although the effects differed depending on the initial arousal state, airflow stimulation to the neck has the potential to maintain driver arousal while suppressing thermal discomfort.

Development of a Drowsiness Assessment and Prediction System (DAPS) for Commercial Vehicle Drivers: A Basic Study on Arousal Level Assessment Methods

Nobuhisa Tanaka･Tetsuya Niikuni (NALTEC)

We conceptualized a Drowsiness Assessment and Prediction System (DAPS) for commercial vehicle drivers to reduce drowsy-driving crashes by detecting declines in arousal during driving after the pre-duty roll call (tenko) and predicting the onset of drowsiness. Focusing on sleepiness-related changes in working memory and speech-related performance, we conducted an overnight, repeated-measures human-subject experiment in which tests assessing these functions were administered repeatedly. Test performance declined as sleepiness increased, demonstrating the promise of the indices used in each test.