Tesla Autopilot Through Constructions: Investigating the Effect of On-Road Partially-Automated Driving through Construction Zones

Biondi, Francesco; Sahoo, Praneet; Jajo, Noor · 2024 · Crossref

DOI: 10.21203/rs.3.rs-4675940/v1

archive: archived pipeline: cataloged

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

Summary

This study investigates the impact of Tesla Autopilot (SAE Level 2 partial automation) on driver cognitive workload and glance allocation when navigating through construction zones, a high-risk environment for collisions. The research addresses a gap in the literature, which has predominantly focused on driver interaction with partial automation in favorable traffic conditions, while ignoring accident-prone areas like work zones where crash risks are significantly elevated. The primary objectives were to determine if driving mode (manual vs. Level 2) and road zone (pre-construction, construction, post-construction) affect cognitive workload and visual attention distribution. The experiment involved 21 participants driving a 2022 Tesla Model 3 on a 68-kilometer section of Ontario Highway 401. The design was a within-subject factorial study where participants drove the route twice: once in manual mode and once in Level 2 mode (with Adaptive Cruise Control and Lane Keeping Assist engaged), with the order counterbalanced. The route was divided into three distinct zones: pre-construction, construction (marked by signs and cones), and post-construction. Cognitive workload was measured using the ISO Detection Response Task (DRT), a vibrotactile reaction time test administered during driving. Glance allocation was tracked via three high-speed cameras (driver view, forward view, touchscreen view) and manually coded to calculate the percentage of time spent looking at four Areas of Interest (AOIs): forward roadway, vehicle touchscreen, side mirrors, and rearview mirror. Data were analyzed using Bayesian factor analyses to evaluate evidence for or against null hypotheses. Results indicated no significant differences in cognitive workload between manual and Level 2 driving modes (BF = 0.19 for mode), nor across the three road zones (BF = 0.18 for zone). This suggests that the increased complexity of the construction zone did not significantly elevate measured cognitive load compared to other zones. However, significant differences were found in glance allocation. Drivers spent significantly less time looking at the forward roadway during Level 2 driving (91.0%) compared to manual driving (96.2%). Conversely, time spent looking at the vehicle’s touchscreen was significantly higher in Level 2 mode (6.14%) than in manual mode (2.79%). Crucially, these glance patterns did not change across zones; drivers did not increase their forward roadway glances when entering the construction zone, nor did they reduce their attention to the touchscreen. The percentage of time looking at the forward roadway remained statistically stable across pre-construction, construction, and post-construction zones. The findings imply that the distraction patterns associated with partial automation persist even in high-risk environments. Drivers failed to adjust their visual attention to account for the heightened hazards of the construction zone, continuing to glance at the touchscreen at the same rate as in safer pre-construction areas. This suggests that the "underload" state induced by Level 2 automation may prevent drivers from self-regulating their attention appropriately when road conditions become more demanding. The study highlights a potential safety risk: the assumption that partial automation improves safety in complex scenarios may be flawed if drivers do not compensate for the system's limitations by increasing vigilance. The authors argue that systems should be engineered to automatically disengage in unfavorable conditions, as human drivers may not reliably detect the need to resume control.

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 qwen3.8-27b-gittensor on 2026-08-23; verification: pending re-verification.

Topics

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

• work zones
• situational awareness
• hands on hands off engagement
• automation surprise
• temporal
• automation

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