AIoT-based Continuous, Contextualized, and Explainable Driving Assessment for Older Adults

Yimeng Liu, Fangwei Zhang, Maolin Gan, Jialuo Du, Jingkai Lin, Yawen Wang, Fei Sun, Honglei Chen, Linda Hill, Ruofeng Liu, Tianxing Li, Zhichao Cao

TL;DR

AURA is an AIoT framework designed to move driving assessment for older adults from infrequent clinic "snapshots" to continuous, real-world monitoring. By fusing multi-scale behavioral modeling with environmental context (like weather and traffic), AURA identifies subtle cognitive declines while preserving privacy through an edge-based, federated learning architecture.

What Problem Does This Paper Address?

As the global population ages, maintaining driving safety is a critical challenge for independence. Current assessment methods—like periodic license renewals or clinical cognitive tests (e.g., MMSE)—are episodic and reactive. They often fail to capture how a senior actually performs in the complex, dynamic environment of real-world roads.

Furthermore, older adults are masters of "self-regulation." They may avoid night driving or complex intersections to compensate for declining abilities. This creates a "personalized driving signature" that can mask underlying functional decline from traditional data-driven models. The authors tackle the challenge of distinguishing between a driver's lifelong habits, their compensatory strategies, and genuine cognitive impairment.

Methods

The authors propose AURA (Aging Understanding & Risk Assessment), a multi-layered AIoT framework:

  • Multi-Scale Representation Learning: The system captures data at different granularities. At the low level, it uses temporal convolutions and self-attention to analyze "micro-movements" (pedal pressure, steering jitter, head-scanning). At the high level, it builds daily and weekly profiles to identify long-term behavioral drift.
  • Context-Integrated Modeling: AURA doesn't look at the driver in a vacuum. It integrates environmental data (traffic density, road geometry, weather) to "normalize" behavior. For example, it determines if a slow reaction was due to a complex intersection or a lapse in attention.
  • Transparent and Actionable Intelligence: Instead of a "black-box" risk score, AURA uses a Semantic Abstraction pipeline. It maps specific driving errors to clinical cognitive domains (e.g., mapping lane-keeping drift to visuospatial processing issues), producing a "Preliminary Diagnostic Report" that clinicians can actually use.
  • Privacy-Preserving Edge Architecture: To handle sensitive location and behavioral data, AURA performs inference locally on in-vehicle hardware. It uses Federated Learning (FL) to update global models without ever transmitting raw sensor data or video to the cloud.

Datasets

The study draws insights from three distinct sources to validate their vision:

  1. CARLA Simulator Testbed: A high-fidelity simulation involving 20 participants (10 seniors, 10 youth) equipped with Azure Kinect for full-body tracking and high-frequency telemetry.
  2. LongROAD: A massive longitudinal dataset of 2,402 older adults tracked over 5 years, providing trip-level logs and mileage records.
  3. DRIVES: A dataset of 414 older adults over 3 months that pairs high-resolution GPS/OBD-II data with formal neuropsychological assessments.

Key Insights and Ideas

  • The "Signature" Problem: Using Principal Component Analysis (PCA), the researchers found that driving data clusters much more strongly by individual identity than by cognitive status. A person's unique driving style is "louder" than the signal of their cognitive decline, making personalized baselines essential.
  • Compensation as a Signal: Seniors exhibit wider and more frequent head-scanning. This isn't just a behavior; it's an adaptive strategy to compensate for reduced peripheral vision and divided attention.
  • Contextual Sensitivity: The presence of a "lead vehicle" significantly changes senior behavior (less throttle, more braking). Without knowing a car was in front of the driver, a model might incorrectly flag this cautiousness as a sign of sluggishness or impairment.
  • The Importance of "Why": A reduced number of trips could mean cognitive decline, or it could just mean the driver retired and stays home more. The paper argues that AI must move toward causal inference to distinguish lifestyle changes from functional loss.

Key Results

  • Risk-Averse Profiles: Seniors consistently demonstrate lower throttle usage, slower acceleration, and more abrupt braking compared to younger drivers in controlled simulations.
  • Mobility Decline: Analysis of the LongROAD data showed that monthly driving distance drops by more than 50% between drivers aged 65–74 and those over 80.
  • Safety Indicators: Using unsupervised clustering, the authors identified that trip frequency, driving days, and short-distance trip ratios are the most dominant indicators of age-related adaptation in mobility patterns.
  • Route-Specific Stress: Seniors showed maximal divergence from younger drivers at intersections (delayed reactions) and curves (larger route deviations).

Limitations and Future Work

The authors acknowledge that while they can detect patterns, they cannot yet fully explain the underlying mechanisms (e.g., is a missed turn due to memory loss or a poorly visible sign?).

Future Directions include: * Elderly Cognitive World Models: Developing AI that explicitly represents how older adults interpret situations and form expectations. * Digital Therapeutics: Moving from monitoring to intervention, where the vehicle provides subtle, context-aware prompts to "exercise" the driver's cognitive functions during low-risk segments of a trip. * Federated Embodied Learning: Refining how vehicles learn from each other's "experiences" on the road without compromising individual driver privacy.

Summary updated

Abstract

The world is undergoing a major demographic shift as older adults become a rapidly growing share of the population, creating new challenges for driving safety. In car-dependent regions such as the United States, driving remains essential for independence, access to services, and social participation. At the same time, aging can introduce gradual changes in vision, attention, reaction time, and driving control that quietly reduce safety. Today's assessment methods rely largely on infrequent clinic visits or simple screening tools, offering only a brief snapshot and failing to reflect how an older adult actually drives on the road. Our work starts from the observation that everyday driving provides a continuous record of functional ability and captures how a driver responds to traffic, navigates complex roads, and manages routine behavior. Leveraging this insight, we propose AURA, an Artificial Intelligence of Things (AIoT) framework for continuous, real-world assessment of driving safety among older adults. AURA integrates richer in-vehicle sensing, multi-scale behavioral modeling, and context-aware analysis to extract detailed indicators of driving performance from routine trips. It organizes fine-grained actions into longer behavioral trajectories and separates age-related performance changes from situational factors such as traffic, road design, or weather. By integrating sensing, modeling, and interpretation within a privacy-preserving edge architecture, AURA provides a foundation for proactive, individualized support that helps older adults drive safely. This paper outlines the design principles, challenges, and research opportunities needed to build reliable, real-world monitoring systems that promote safer aging behind the wheel.
BibTeX
@misc{liu2026aiot,
  author = {Yimeng Liu and Fangwei Zhang and Maolin Gan and Jialuo Du and Jingkai Lin and Yawen Wang and Fei Sun and Honglei Chen and Linda Hill and Ruofeng Liu and Tianxing Li and Zhichao Cao},
  title = {{AIoT-based Continuous, Contextualized, and Explainable Driving Assessment for Older Adults}},
  year = {2026},
  eprint = {2603.00691},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2603.00691},
}
Figure