Is partially automated driving a bad idea? Observations from an on-road study

Banks, Victoria A.; Eriksson, Alexander; O'Donoghue, Jim; Stanton, Neville A. · 2018 · Crossref

DOI: 10.1016/j.apergo.2017.11.010

archive: archived pipeline: cataloged verified

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

Summary

This study investigates the safety implications of partially automated driving systems, specifically SAE Level 2 features like Tesla’s Autopilot, which automate longitudinal and lateral control while requiring the driver to remain in an active monitoring role. The research is motivated by the inherent conflict between system design and human capabilities: while these systems free the driver’s hands and feet, they demand sustained vigilance, a task at which humans are notoriously inefficient. The authors question whether such partial automation appropriately supports drivers or instead fosters complacency, over-trust, and risky behavior that compromises safety. To address this, the researchers conducted an on-road study using a right-hand drive Tesla Model S equipped with Autopilot version 7.x. Twelve participants, all experienced with Advanced Driver Assistance Systems, drove on public roads in Warwickshire, UK, for approximately 40 minutes each. The study utilized a naturalistic approach, with participants encouraged to drive comfortably without active encouragement to remove hands from the wheel, though they were reminded of their legal responsibility for vehicle safety. Data was collected via four synchronized video cameras capturing the driver, the Human-Machine Interface (HMI), and the road environment. A qualified safety driver was present to intervene if necessary. The video data underwent thematic analysis to identify patterns in driver behavior, system warnings, and mode transitions. The analysis revealed four primary themes indicating significant safety risks. First, system warnings were frequent; 11 of 12 participants drove "hands-free" for over 60 seconds, triggering visual warnings, with seven cases escalating to auditory alerts after 75 seconds. This indicates that drivers often failed to monitor the HMI, leading to delayed responses. Second, drivers exhibited mode confusion, mistakenly believing the system was engaged when it was not, or failing to notice inadvertent deactivation. Third, two participants intentionally tested the boundaries of the system’s Operational Design Domain (ODD), engaging in risky behavior while assuming the system would handle the maneuver. Finally, drivers engaged in non-driving related secondary tasks, such as drinking coffee or turning away from the road, demonstrating a shift from a monitoring role to a "Driver Not Driving" state. These behaviors suggest that drivers quickly become complacent and over-trust the automation, losing situational awareness. The study concludes that partially automated driving poses significant risks because it relies on humans to perform sustained monitoring tasks they are ill-equipped to handle. The authors argue that the current design ethos creates an "impossible task" for drivers, leading to inevitable errors that are often misattributed to driver failure rather than design flaws. They suggest that SAE Level 2 and 3 systems are fundamentally flawed because they place the human as the last line of defense in a system where vigilance decays rapidly. The authors recommend that future automation should either keep the driver in full control (Level 1) or move directly to higher levels of automation (Level 4) where the human is removed from the control loop entirely, thereby eliminating the dangerous intermediate monitoring role.

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.

Summary generated by qwen3.6-27b-nvidia on 2026-08-10; verification: verified.

Topics

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

• automation
• automation surprise
• hands on hands off engagement
• mode awareness
• automation complacency bias
• situational awareness

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).

• Empirical Findings: behavioral performance data
• Theoretical Contribution: conceptual framework, theory or model