Autonomous Vehicle Dataset with Real Multi-Driver Scenes and Biometric Data.

Area of Science:

• Robotics
• Computer Vision
• Human-Computer Interaction

Background:

• Autonomous vehicle development is rapidly advancing, with a growing need for comprehensive real-world data.
• Existing datasets often lack crucial human interaction data, hindering the evaluation of higher autonomy levels.
• Synthetic datasets are gaining traction due to cost and availability but may not fully represent real-world complexities.

Purpose of the Study:

• To introduce the UPCT dataset, a novel public resource for autonomous vehicle research.
• To provide high-quality, multimodal data encompassing sensor and human factors.
• To make associated data synchronization and processing software publicly available.

Main Methods:

• Collected multimodal data using state-of-the-art sensors (3D LiDAR, cameras, IMU, GPS, encoders) onboard the UPCT CICar autonomous vehicle.
• Integrated driver biometric data and driver behavior questionnaires.
• Developed and released software for data synchronization and processing.

Main Results:

• The UPCT dataset contains diverse, high-quality sensor and human interaction data.
• An end-to-end neural network model utilizing the dataset achieved promising results in predicting speed and steering angle.
• The dataset and associated software are publicly available for research.

Conclusions:

• The UPCT dataset addresses a critical gap by including human interaction data, essential for advanced autonomous systems.
• The dataset's quality and the promising validation results support its utility for training and evaluating autonomous driving models.
• Public availability of the dataset and software will accelerate research and development in the autonomous vehicle sector.