Exploring AI-Driven Approaches to Quantify and Mitigate Driver Distraction

As automated systems and assistive driver features continue to advance, driver distraction is increasing in both manual and semi-autonomous modes. Automated vehicles (AVs) offer promising technology solutions to reduce distraction risk. However, automation can create out-of-the-loop periods, encouraging non-driving tasks and masking gradual disengagement. Current AVs are semi-autonomous, so drivers must be ready to retake control for issues like missing lane markers. The takeover process is cognitively and physically demanding, unfolding within seconds, which is especially challenging when drivers are distracted. Traditionally, distractions in vehicles have mainly been due to phone use or infotainment systems. New technologies have expanded the sources of distraction. These distractions include visual distractions where drivers’ eyes are diverted from the road (e.g., using a GPS), manual distractions where drivers remove their hands from the wheel (e.g., using a phone or eating while driving), and cognitive distractions where a driver’s mind is not focused on driving (e.g., mind wandering). Emotions and fatigue that pull attention from driving also reduce engagement. Social pressures and smartphone habits further contribute to distraction. This emphasizes the need to broaden the definition of distraction to include driver engagement levels to better reflect real-time mental states and driving performance. Given the many types of distractions, there is a need to systematically quantify their effects using both driving and physiological measures acquired by various sensors under a more universal paradigm (e.g., engagement level). With the rapid advancements in artificial intelligence (AI), particularly in the domain of deep learning, sophisticated multimodal models have emerged as powerful tools for integrating diverse data sources. These models offer an unparalleled ability to combine acquired features from various sensors into a cohesive understanding of driver states. Leveraging such techniques enables us to capture nuanced indicators of distraction, which can ultimately enhance the system’s ability to predict and respond to varying levels of engagement. The research team proposes using federated learning with multimodal sensor fusion, where each sensor model is trained locally on-device to maintain privacy and adapt to specific driving conditions. Federated learning enables these models to share knowledge without centralizing data, creating a global model that can recognize diverse distractions across different driving contexts. Therefore, the goals of this project have two parts: (1) Identify and quantify the effects of traditional and emerging distractions, and (2) Build an AI-driven framework that categorizes and quantifies all types of driver distractions based on multisensory data. Two phases of studies will be conducted to achieve the two goals, respectively. Phase I will identify and quantify both traditional and emerging forms of driver distraction. Specifically, the team will first conduct a national survey to better assess sentiments, attitudes, and key themes regarding driver distraction as well as correlations and trends among demographics, distraction types, and self-reported behaviors. With the knowledge gained, the team will then quantify the effects of distractions on driving performance and physiological measurements. The team will categorize types of distractions based on established classifications such as visual, manual, and cognitive distractions, as well as emerging types identified in the survey. Then the team will conduct an in-lab experiment using a high-resolution driving simulator. Participants’ driving and physiological measures. All data collected herein will then be used in Phase II, the development of an AI-driven framework. Phase II will develop an AI-driven framework to categorize distraction types using multisensory data. The team will integrate three sensors, eye tracking, a depth sensor, and face video, to enable a complete view of distraction. Multimodal models will fuse these signals to improve accuracy and robustness. The team will aggregate insights from each modality to achieve a more reliable categorization of distraction types. Federated learning will build a global model while keeping data on the device, which supports privacy and adaptation to different platforms and roads. The global model will serve as the core engine in Phase II, classifying distraction types based on combined sensory inputs. This layered AI approach will strengthen the understanding of driver distraction and support real-time interventions.

Language

• English

Project

• Status: Active
• Funding: $109,708.00
• Contract Numbers:

  69A3552348323

• Sponsor Organizations:

  Office of the Assistant Secretary for Research and Technology

  University Transportation Centers Program
  Department of Transportation
  Washington, DC  United States  20590
• Managing Organizations:

  Howard University

  2400 6th Street, NW
  Washington, DC  United States  20059
• Project Managers:

  Bruner, Britain

• Performing Organizations:

  San Jose State University

  1 Washington Sq
  San Jose, California  United States  95192
• Principal Investigators:

  Huang, Gaojian

• Start Date: 20251201
• Expected Completion Date: 20261126
• Actual Completion Date: 0
• USDOT Program: University Transportation Centers Program

Subject/Index Terms

• TRT Terms: Artificial intelligence; Autonomous vehicle handover; Distracted driving; Distraction; Driving simulators; Sensors
• Subject Areas: Highways; Safety and Human Factors; Vehicles and Equipment;

Filing Info

• Accession Number: 01976551
• Record Type: Research project
• Source Agency: Research and Education for Promoting Safety (REPS) University Transportation Center
• Contract Numbers: 69A3552348323
• Files: UTC, RIP
• Created Date: Jan 19 2026 4:14PM