Driver-initiated Tesla Autopilot Disengagements in Naturalistic Driving

Morando, Alberto; Gershon, Pnina; Mehler, Bruce; Reimer, Bryan · 2020 · Crossref

DOI: 10.1145/3409120.3410644

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 driver behavior during the transition from Tesla Autopilot (AP) to manual control in naturalistic driving conditions. Motivated by limited objective data on how drivers interact with Level 2 automation in real-world traffic, the research aims to quantify changes in visual attention and steering wheel control. The authors seek to understand whether drivers maintain adequate supervision while using AP and how they re-engage when disengaging the system, addressing concerns about over-reliance and reduced situational awareness. The researchers utilized data from the MIT Advanced Vehicle Technology (MIT-AVT) naturalistic driving study, which collected continuous video and vehicle telemetry from Tesla Model S and X owners. From a large dataset, they selected 298 non-critical, driver-initiated AP disengagements on highways involving 19 drivers. For each event, they analyzed a 30-second segment (20 seconds before and 10 seconds after disengagement). Glance behavior was coded frame-by-frame into categories such as road, instrument cluster, and center stack. Steering wheel control was categorized into four levels: high (both hands at 3-9 o’clock), medium, low, and none (no hands on the wheel). The results indicate significant differences in behavior between AP use and manual driving. During steady-state AP use, drivers directed 64% of their glances to the road, compared to 76% during manual driving. Conversely, glances to the center stack increased from 8% in manual driving to 17% with AP active. Regarding steering, 33% of drivers had no hands on the wheel while AP was engaged, a proportion that dropped to 1% after disengagement. During the transition phase, drivers progressively increased on-road glances and moved their hands to the steering wheel. By the time of disengagement, on-road glances rose to 82%, and the proportion of drivers with no hands on the wheel fell to 5%. The findings suggest that Tesla AP users frequently divert attention from the road, particularly toward the center stack, and often remove their hands from the steering wheel while the system is active. However, drivers appear to anticipate planned disengagements, gradually re-engaging visually and physically before taking control. The study highlights that while AP reduces immediate driving demand, it may lead to complacent reliance. The authors conclude that these real-world insights are crucial for designing safer assistive systems and establishing appropriate standards for driver monitoring and engagement.

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.

• hands on hands off engagement
• automation
• manual
• takeover transitions
• automation surprise
• temporal

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
• Methodological Resource: tool software
• Theoretical Contribution: conceptual framework