MIT Advanced Vehicle Technology Study: Large-Scale Naturalistic Driving Study of Driver Behavior and Interaction With Automation

Fridman, Lex; Brown, Daniel E.; Glazer, Michael; Angell, William; Dodd, Spencer; Jenik, Benedikt; Terwilliger, Jack; Patsekin, Aleksandr; Kindelsberger, Julia; Ding, Li; Seaman, Sean; Mehler, Alea; Sipperley, Andrew; Pettinato, Anthony; Seppelt, Bobbie; Angell, Linda; Mehler, Bruce; Reimer, Bryan · 2019 · IEEE Access

DOI: 10.1109/access.2019.2926040

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

The MIT Advanced Vehicle Technology (MIT-AVT) study addresses the critical gap in understanding human interaction with semi-autonomous driving systems. While previous research focused on crash epochs or controlled experiments, this study argues that full autonomy remains unsolved due to complex real-world edge cases and human variability. Consequently, humans will remain integral to the driving task for decades, monitoring AI systems that perform varying percentages of the driving. The study aims to characterize how drivers engage with automation in naturalistic settings, understand human-machine interaction dynamics, and inform the design of shared autonomy systems that enhance safety. To achieve these objectives, the researchers conducted a large-scale naturalistic driving study using 29 instrumented vehicles, including Tesla Model S/X, Volvo S90, Range Rover Evoque, and Cadillac CT6 models. The fleet included both subject-owned vehicles for long-term monitoring (over one year) and MIT-owned vehicles for medium-term observation (one month). Data collection involved 122 participants, accumulating 511,638 miles, 15,610 participant days, and 7.1 billion video frames. The instrumentation captured high-definition video streams of the driver’s face, cabin, forward roadway, and instrument cluster, alongside GPS, IMU, and CAN bus telemetry. This multi-modal approach allows for the analysis of the "long-tail" of driving behavior rather than just isolated crash events. The study employs deep learning-based computer vision algorithms to automate the extraction of actionable knowledge from the massive dataset, overcoming the limitations of manual annotation. Specific analytical tasks include fine-grained face recognition for gaze and expression detection, body pose estimation to monitor driver alertness and position, semantic scene perception for environmental understanding, and vehicle state detection from instrument cluster video. By integrating these automated visual metrics with driver questionnaires regarding mental models, trust, and self-reported experiences, the study creates a holistic view of real-world technology use. This methodology enables the identification of patterns in how drivers allocate attention and interact with automation features like Autopilot over extended periods. The significance of the MIT-AVT study lies in its establishment of a new standard for naturalistic driving research. By leveraging deep learning to analyze billions of frames, the study provides unprecedented insight into the continuous, moment-to-moment dynamics of shared autonomy. This data supports the development of Human-Centered Artificial Intelligence (HCAI) systems, aiming to prevent unintended consequences in human-AI interaction. The findings are intended to guide the design of safer vehicle systems, inform insurance providers about changing safety markets, and educate policymakers on the realities of automation adoption. The ongoing nature of the study ensures a growing dataset that reflects evolving driver behaviors and technology updates.

Key finding

At publication the MIT-AVT dataset comprised 122 participants, 15,610 participation days, 511,638 miles, and 7.1 billion HD frames from 29 instrumented semi-autonomous vehicles, establishing a deep-learning-enabled naturalistic platform for analyzing long-tail driver–automation interaction beyond traditional crash-epoch annotation.

Methodology

naturalistic

Sample size: 122

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed. Discovered via openalex_abstract on 2026-05-08 (3 acquisition events logged).

Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• situational awareness
• naturalistic crash near crash
• automation surprise
• automation
• distraction detection algorithms
• exposure measurement

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Methodological Resource: dataset resource, tool software
• Theoretical Contribution: computational model