Keywords

1 Introduction

The Automated Lane Keeping System (ALKS) was the first Automated Driving System (ADS) application to have been regulated by the United Nations Economic Commission for Europe (UNECE) via Regulation 157 (UN-R157). ALKS is a system enabling automated driving on motorways with physical separation among opposite driving lanes at a cruising speed of up to \(60~\mathrm {km/h}\). The latest amendment increased the operational speed up to \(130~\mathrm {km/h}\) and introduced automatic lane change [5]. A stand-out feature of any ADS feature with respect to Advanced Driver Assistance Systems (ADAS) is that the driver is no longer responsible for the driving task when the system is engaged. Thus, an ADS achieves automation Level 3 or 4 according to SAE J3016 [4]. At the time of writing, only two vehicles exist in Europe that were type-approved according to the ALKS provisions: the Mercedes S-Class and Mercedes EQS.

In terms of user-perceived functionality and target operational domain, the ALKS L3 automation is in direct competition with the Adaptive Cruise Control (ACC) L2 assistance. In contrast to ALKS, ACCs are envisioned as SAE J3016 Level 1/2 systems meaning that the driver shall always be ready to take back control and the same will remain legally liable. ACCs have reached nowadays a substantial market penetration which eases carrying out testing to understand real-world behavior. Despite the undoubted comfort benefit that ACC can deliver, several concerns have been raised in the literature about the traffic implications [1]. For instance, the empirical evidence suggests that ACCs exhibit poor string stability performances and slow reaction times thus contradicting many fundamental hypotheses which advocated for the introduction of such systems.

To the end of providing an empirically-informed study that could give practitioners more realistic assumptions and monitor the implementation of the legal provisions, the authors have organized a testing campaign in Germany involving one of those ALKS-featured vehicles. The objective was to provide a quantitative assessment of the system’s capabilities with respect to the competing ACC. Due to the limited Operational Design Domain (ODD) of the ALKS feature (the manufacturer according to UN-R 157 is free to select the ODD), the authors could only test the system in a selection of German motorways where the following conditions applied simultaneously: no roadworks, no tunnel, good weather, good visibility, and traffic speed up to \(60~\mathrm {km/h}\) due to congestion.

This paper provides the preliminary findings of a testing campaign aimed at assessing the real impact of driving automation on public roads. The first results suggest that ALKS show remarkably better string stability performance when traveling under comparable headway in relation to the traditional Adaptive Cruise Control (ACC) system.

2 Methodology

2.1 Testing Campaign

The experimental campaign took place in November 2023, about 3000 km were driven in Germany using 2 vehicles equipped with state-of-the-art ADAS and a vehicle equipped with ALKS from different OEMs while trying to maximize the use of the driving assistance/automation features. The vehicles were instrumented with external GNSS antennas to record the positions and velocities to enable the later post-processing vehicle-independent analysis.

2.2 String Stability

String stability refers to the capability of a platoon to absorb a traffic perturbation. Conversely, a string stable platoon will dampen disturbances whereas a string unstable platoon will magnify the same leading to potentially unsafe driving scenarios. String stability, \(w_{SS}\), is given by

$$\begin{aligned} w_{SS} = \frac{v_{FF,leader}-\min (v_{follow})}{v_{FF,leader}-\min (v_{leader})}, \end{aligned}$$
(1)

where \(v_{FF,leader}\) is the free-flow velocity of the leader before the perturbation occurs, \(\min (v_{follow})\) is the minimum velocity recorded for the follower, and \(\min (v_{leader})\) is the minimum velocity recorded for the leader. \(w_{SS}\le 1\) indicates a string stable platoon.

2.3 Headway Policy

The headway refers to the spacing policy adopted by a follower vehicle when cruising at equilibrium in car-following mode. The headway policy and string stability have been demonstrated to exist in a trade-off interconnection. On one side, a low spacing policy enables high theoretical flow. However, the poor string stability associated with such reduced distances will in practice deteriorate the flow as soon as a perturbation occurs or due to the intrinsic instability of the system. On the other hand, increasing the headway will make the platoon more robust against traffic perturbations but at the expense of reduced theoretical maximum flow. As such, the quantification of the string stability cannot be decoupled from the corresponding headway characterization.

In our testing campaign, we included in the dataset long portions of constant speed driving segments to estimate at the post-processing phase the corresponding headway settings. From the collected evidence the time-gap \(t_g\), which together with the length of the vehicle originates the headway, can be estimated as

$$\begin{aligned} t_g = \dfrac{dist_{\text {long}}}{v_{follow}}, \end{aligned}$$
(2)

where \(dist_{\text {long}}\) is the bumper-to-bumper longitudinal distance at equilibrium.

2.4 Reaction Time

The reaction time is a relevant metrics for the safety assessment of a driving automation technology. In the following, the definition suggested in [3] is adopted where the reaction time \(t_r\) is the time offset that maximizes the Pearson correlation \(r(\cdot )\) between the speed difference of two adjacent traffic participants (\(\Delta v\)) and the follower’s produced acceleration (\(a_{follow}\)) in response to a perturbation as of

$$\begin{aligned} t_r = \arg \max (r_{\Delta v, a_{follow}}(\Delta v(t), a_{follow}(t+t_r))). \end{aligned}$$
(3)

3 Results

3.1 String Stability

Figure 1 depicts, on the left, the string unstable behavior denoted for the ACC system whereas, on the right, the completely different behavior reported when the same vehicle traveled in ALKS mode is shown. The behavior illustrated for one specific scenario in Fig. 1 has been consistent throughout the several perturbations recorded in the testing campaign.

Fig. 1.
figure 1

ACC (left) string unstable behavior vs. ALKS (right) string stable behavior.

Full size image

By aggregating the string stability metrics computed per each disturbance scenario reported, Fig. 2 can be obtained which shows the median values and the corresponding Q1/Q3 quartiles. Overall, significantly better performances can be denoted for the ALKS since the entire distribution is contained in the region \(w_{SS} < 1\) suggesting that the system has always been string stable in the dataset collected. The same considerations do not apply to the ACC where the distribution is always above the string stability threshold. Concerning the numerical assessment, the median \(w_{SS}\) for the ALKS is 0.83 with an Inter Quartile Range (IQR) of 0.14. Conversely, the ACC displays a median \(w_{SS}\) of 1.31 and an IQR of 0.20.

Fig. 2.
figure 2

ACC vs. ALKS aggregated string stability margins.

Full size image

3.2 Headway Policy

Figure 3 depicts the aggregated data concerning the time-gap \(t_g\) for the ACC and ALKS. The numerical assessment returns a median \(t_g\) for the ALKS equal to \(1.60~\textrm{s}\) with an IQR of \(0.07~\textrm{s}\). Conversely, the ACC displays a median \(t_g\) of \(1.50~\textrm{s}\) and an IQR deviation of \(0.41~\textrm{s}\).

Fig. 3.
figure 3

ACC vs. ALKS aggregated headway policies.

Full size image

The ACC has a slightly lower median headway (\(\approx 6\%\)) than the ALKS which would imply a theoretically higher maximum flow. However, the poor string stability performance as of Fig. 2 while operating such a short time-gap will most likely annihilate any potential theoretical advantage in traffic flow in a real-world scenario. Additionally, the ALKS has a minimum time-gap of \(1.60~\textrm{s}\) when traveling at \(60\mathrm {km/h}\) as mandated by UN-R157, a provision which is proven to be fulfilled. Eventually, the ACC also exhibits a substantially larger dispersion that can be ascribed to the mentioned poorer stability performance and generally less precise tracking of the leader’s speed.

3.3 Reaction Time

Figure 4 depicts the aggregated data concerning the reaction time for the ACC and ALKS estimated using (3). The same perturbations used to identify the string stability margin were adopted to compute the aggregate reaction time. The median reaction time for the ALKS has been identified as \(0.80~\textrm{s}\) with an IQR of \(0.20~\textrm{s}\). The ACC is significantly slower with a median \(r_t\) of \(2.60~\textrm{s}\) and IQR of \(0.90~\textrm{s}\).

Fig. 4.
figure 4

ACC vs. ALKS aggregated reaction time.

Full size image

The results show the clearly different nature of ALKS vs ACC. The automation system is tuned to be more reactive to traffic disturbance since it is not only designed for comfort applications, in contrast to its driving assistance counterpart, but for tackling safety-critical scenarios as it bears the driving responsibility.

4 Conclusion

The study summarized the results of the first commercially available ALKS system in Europe. To the best of the authors’ knowledge, this scientific effort is the first independently assessed documented evidence regarding commercial automation solutions moving beyond SAE J3016 Level 2.

The preliminary outcomes look encouraging. The ALKS managed to substantially improve the string stability metrics with negligible impact on the headway policy thus contributing to increased real-world flow and potentially better fuel consumption. Additionally, the legal provisions mandating a minimum level of safety culminate in a safety less prone to delayed reaction.

Still, the very limited ODD remains the main challenge OEMs will have to face in order for the benefits to become tangible to the wider transportation network. Further work will be devoted to extending the performance comparison to embrace also previously tested ACC systems in the past [2] and to release publicly the data to the wider practitioners’ community.