Silence Is Not Assurance: When Production Lane Keeping Assist Does Not Warn, and What It Costs
Authors: Tianjia Yang, Yuxin Ding, Ke Li, Kaidi Liang, Ruwen Qin, Xianbiao Hu
Organizations: Department of Civil and Environmental Engineering, The Pennsylvania State University, University Park, PA 16802, USA · Department of Civil Engineering, Stony Brook University, Stony Brook, NY 11794, USA
Production lane keeping assist (LKA) systems tell the driver whether they detect the lane markings, but whether this displayed status reflects what the system perceives, and what happens when it does not, has not been examined in real-world driving. This study introduces CANDID-LKA, a CAN-based Diagnosis of the Indicated Detection status in LKA. It evaluates a proprietary driver-facing status using three signals: the displayed status and the steering torque of the driver, both recorded from the controller area network (CAN) bus of the vehicle, and the lane perception confidence of a third-party vision system observing the same road. The framework examines how the status responds to driving conditions and estimates how completely it reports degraded lane perception by comparing it with the other two signals in a latent-class model. We apply it to 39,800 seconds of real-world driving from 13 vehicle models of three manufacturers. The results show that the status does not respond consistently to the conditions that make markings harder to see, and that the three makes differ widely in how often they report not detected. The latent-class model shows that the displayed status is incomplete: it reports only 47% of degraded seconds as not detected, compared with 80% for the third-party low perception confidence, and in half of the low-confidence episodes it never reports not detected. In the silent seconds in which the status reads detected while the perception confidence is low and the driver is not in control, the standard deviation of lateral position over the next 3 s is about 60% larger than when the status and the perception confidence agree. A detected status thus does not guarantee reliable lane perception, and silence with the driver not in control is the case that most needs attention.
Figures & tables
Symbol
Variable
Definition
Source
Ot
Displayed status
1 when the vehicle reports the lane as not detected , 0 when it reports detected
Vehicle
Ct
Lane perception confidence
1 when the third-party vision system’s perception confidence for the lane line is low (below 0.5), 0 otherwise
Third-party vision system
It
Driver control response
1 when the driver’s steering torque exceeds the vehicle’s 95th percentile within the following 3 s
Vehicle
Ht
Driver in control
1 when the driver’s mean steering torque over the following 3 s exceeds the vehicle’s median per-second torque
Vehicle
Yt
Lateral position
Position of the vehicle in the lane estimated by the vision system at 10 Hz; the 3 s SDLP and lateral excursion are computed from it
Third-party vision system
Xt
Driving conditions
Marking condition (clear, low-contrast, faded, or occluded), lighting, surface, road segment, traffic, lead-vehicle presence, speed, and road curvature
Scene labels from the video; vehicle; vision system
Table 1: Variables used by CANDID-LKA for each driving second t
Displayed status ( not detected )
Perception
Condition
All systems
Toyota
Hyundai
Honda
confidence (low)
Marking condition (vs. clear)
Low-contrast
0.79 (0.48–1.29)
0.81 (0.42–1.58)
0.67 (0.29–1.56)
1.92 (1.16–3.16) ∗
1.31 (0.98–1.75)
Faded
0.99 (0.53–1.87)
0.55 (0.30–1.00)
2.11 (1.07–4.16) ∗
2.60 (1.17–5.78) ∗
1.31 (0.79–2.19)
Occluded
1.94 (0.79–4.76)
3.54 (1.09–11.42) ∗
1.54 (0.34–6.92)
0.38 (0.23–0.62) ∗
0.82 (0.34–2.00)
Other conditions
Table 2: Odds ratios for a not-detected report under different driving conditions
Transition
Low-contrast
Faded
Occluded
All systems
Loss
1.44 (0.94–2.21)
1.47 (0.85–2.54)
4.10 (1.95–8.60) ∗
Recovery
1.12 (0.55–2.27)
1.18 (0.59–2.39)
1.84 (0.78–4.33)
Toyota
Loss
1.01 (0.54–1.87)
0.95 (0.43–2.06)
8.10 (3.73–17.60) ∗
Recovery
1.04 (0.29–3.69)
1.57 (0.89–2.78)
2.77 (0.92–8.31)
Hyundai
Loss
1.51 (0.72–3.15)
1.50 (0.54–4.20)
3.61 (1.75–7.44) ∗
Recovery
0.87 (0.40–1.90)
0.46 (0.13–1.70)
1.44 (0.35–5.84)
Table 3: Loss and recovery rate ratios of the displayed status by marking condition
All systems
Toyota
Hyundai
Honda
Prevalence of degraded perceivability
0.13 (0.08–0.19)
0.14 (0.09–0.22)
0.14 (0.07–0.29)
0.05 (0.01–0.13)
Displayed status
Sensitivity
0.47 (0.34–0.62)
0.46 (0.20–0.67)
0.44 (0.33–0.63)
0.79 (0.75–0.89)
Specificity
0.84 (0.74–0.92)
0.75 (0.56–0.92)
0.89 (0.83–0.94)
0.98 (0.94–0.99)
Positive predictive value
0.30 (0.22–0.44)
0.24 (0.15–0.33)
0.40 (0.23–0.67)
0.67 (0.50–0.84)
Negative predictive value
0.92 (0.85–0.95)
0.89 (0.80–0.94)
0.91 (0.79–0.97)
0.99 (0.97–1.00)
Table 4: Latent-class estimates of prevalence, sensitivity, and specificity, by make
Figure 1: Agreement, episode miss rate, and driver-response risk ratio, by make (route-bootstrap 95% CI)
Silent seconds
Agreement seconds
In control
Not in control
In control
Not in control
Share of seconds
78%
22%
48%
52%
Seconds per driving hour
166
48
Median speed (km/h)
29
36
88
92
Maneuver: sharp turn
65%
19%
19%
4%
gentle curve
16%
16%
22%
13%
Table 5: Silent and agreement seconds by whether the driver is in control over the following 3 s, all systems
Silent
3 s SDLP, median (cm)
SDLP adjusted
Excursion above 50 cm
Condition
seconds
agreement / silent
difference (cm)
agreement / silent
All not-in-control seconds
500
4.2 / 12.2
5.4 (3.8–7.0)
6% / 46%
Straight road only
322
3.8 / 10.7
6.4 (5.0–7.8)
3% / 36%
Clear markings
187
4.1 / 11.8
5.5 (3.6–7.3)
6% / 44%
Low-contrast markings
154
4.6 / 11.9
6.3 (3.5–9.0)
6% / 44%
Faded markings
101
4.1 / 13.9
4.1 (0.8–7.4)
8% / 47%
Table 6: Silence with the driver not in control: the 3 s SDLP and lateral excursion that follow, by driving condition, all systems
Analysis
Specification
All systems
Toyota
Hyundai
Honda
Main specification
Threshold 0.5, every second,
0.47
0.46
0.44
0.79
3 s window, independent errors
Perception confidence
0.3
0.50
0.48
0.48
0.86
threshold
0.4
0.49
0.48
0.46
0.81
0.6
0.46
0.47
0.41
0.78
0.7
0.45
0.46
0.39
0.71
Table 7: Sensitivity of the displayed status under alternative analysis choices, by make
Testing of commercial Advanced Driver Assistance Systems is essential to ensure safety and compliance during type approval and in service operation. However, proving ground scenarios may not reflect real world driving complexity, while geo fencing can require manufacturer collaboration and limit assessment independence. This work presents a methodology for independently testing Assisted Lane Change systems on public roads. A campaign on the A31 French motorway used a test vehicle equipped with a LiDAR based vehicle detection and tracking system. Tests covered combinations of inter vehicle distance and speed between the test vehicle and the take over vehicle. Real time kinematic global navigation satellite system receivers assessed detection and tracking performance. Recorded lane change trajectories were compared with the lane change suppression requirements of UNECE Regulation Number 79. Of 27 predefined lane change manoeuvres, 18 were completed and 9 suppressed. In 6 cases, the system allowed manoeuvres that did not meet regulatory minimum distance requirements. In 3 cases, the deviation remained statistically significant after accounting for measurement uncertainty. To the authors knowledge, this is the first public road campaign designed to assess Assisted Lane Change compliance with Regulation Number 79 safety distance requirements. The results demonstrate the suitability of LiDAR based sensing for this purpose. The methodology can support market surveillance and future regulatory revisions by revealing real world behaviours not covered by approval procedures.
Marcello Cellina, Akos Kriston, Antonio Migneco +7
Autonomous driving requires understanding the road as a graph of drivable lanes and their connectivity, beyond the ego lane alone, to follow routes through intersections and reason about cross- and merging-traffic. Recent perception models infer such lane topology, i.e., lane segments together with their inter-lane connectivity (LCLC), from onboard sensors over a 360-degree BEV view. Due to neural perception's imperfections, their outputs retain structural instabilities such as missed detections, lost or incorrect LCLC, over-detection, and label flicker. This paper presents WASABI, a real-time post-processing pipeline that stabilizes lane topology outputs both within and across frames by treating lane segments and their LCLC connectivity as joint tracking targets, under onboard real-time constraints (10 Hz / 20 ms / up to 200 input lanes). The pipeline integrates segment tracking with connectivity, noise-robust topology-aware refinement, and a resource-constrained real-time design. On internal validation data (16 sequences), WASABI improves LCLC detection F1 from 0.834 to 0.948 (+0.114, +13.6%) and reduces centerline lateral error from 2.50 m to 0.95 m, while reducing detection false-positives by 24.6%. Temporal-stability metrics on the same data show LCLC toggle rate reduced by 63.3% and boundary-label flicker rate by 30.2%, confirming across-frame stabilization beyond per-frame accuracy.
Runtime failure monitors can use a model's internal representations to anticipate failures. We audit this monitoring strategy across two autonomous-driving tasks: online vectorized map generation with LaneSegNet and end-to-end planning with VAD. We find that frame-level errors are predictable at inference in both tasks. For LaneSegNet, a supervised latent probe reaches Area Under the Receiver Operating Characteristic curve (AUROC) 0.780 for high Chamfer error; to our knowledge, this is the first post-hoc frame-level failure monitor for online vectorized map generation. For VAD, a supervised planning-latent probe reaches AUROC 0.868 for mean-ADE failure. Our audit shows that internal access is not necessary for strong failure prediction. A monitor using only LaneSegNet's prediction outputs reaches AUROC 0.825, while for VAD, ego state, driving command, and the planner's predicted trajectory reach 0.924 on the same mean-ADE endpoint. Adding latent features to either baseline yields no statistically resolved improvement. This observation persists across a broad suite of planning failure endpoints, including endpoints whose labels depend on geometry unavailable to the non-latent baseline. Thus, predicting failure from an internal representation does not establish that the representation provides useful information beyond observable inputs and outputs. We propose an evaluation protocol for testing the incremental value of latent access and release our per-frame failure endpoint labels.