研究目的
To propose a novel method for determining the utterance timing of a driving agent on the basis of the double articulation analyzer (DAA), and verify the effectiveness of the method through two experiments.
研究成果
The proposed method, based on the double articulation analyzer (DAA), can determine better utterance timing for a driving agent than baseline methods, as shown by the experimental results. However, integrating methods that consider the cognitive load related to each driving word could further improve the performance.
研究不足
The experiments were conducted in an offline manner and in virtual driving situations, limiting the generalizability to real-world driving scenarios. Additionally, the proposed method does not account for the cognitive load related to each driving word, which could improve the accuracy of determining utterance timings.
1：Experimental Design and Method Selection:
The study employs a double articulation analyzer (DAA), an unsupervised nonparametric Bayesian machine learning method for detecting contextual change points in driving behavior data.
2：Sample Selection and Data Sources:
Driving behavior data were collected from test drivers driving a car for five laps of a circuit in Kariya city, Aichi, Japan.
3：List of Experimental Equipment and Materials:
A desktop PC screen and a steering controller (Logicool Driving Force GT) were used for creating a simple virtual driving environment.
4：Experimental Procedures and Operational Workflow:
Participants were asked to virtually drive the car under three conditions corresponding to three methods for determining utterance timing. The driving agent provided non-driving-related information under each condition, and participants' responses were recorded.
5：Data Analysis Methods:
The effectiveness of the proposed method was evaluated through subjective evaluation experiments, comparing it with baseline methods.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
