Platoon of SAE Level-2 Automated Vehicles on Public Roads: Setup, Traffic Interactions, and Stability

Abstract

An increasing amount of vehicles are equipped with driver assistance systems; many of the vehicles currently on the market can be optionally equipped with adaptive cruise control and lane centering systems. Using both systems at the same time brings the vehicle to SAE level-2 automation . This means a driver does not need to perform longitudinal and lateral operational driving, although the driver should be ready to intervene at any time. While this can provide comfort, the interaction between vehicles operated by these systems might cause some undesired effects. This becomes particularly relevant with increasing market penetration rates. This paper describes an experiment with seven SAE level-2 vehicles driven as a platoon on public roads for a trip of almost 500 km. The paper discusses how the experiment was organized and the equipment of the vehicles. It also discusses the interaction of the platoon in traffic, as well as, in basic terms, the interaction between the automated vehicles. The experiences can be useful for other studies setting up field tests. The conclusion from this platoon test is: intentionally creating platoons on public roads is difficult in busy traffic conditions. Moreover, interactions between the vehicles in the platoon show that the current SAE level-2 systems are not suitable for driving as platoons of more than typically three to four vehicles, because of instabilities in the car-following behavior.

Online Parameter Estimation Methods for Adaptive Cruise Control Systems

“…With the estimated parameters from each of the methods, we check the string stability of the vehicle under ACC control. Like all previous studies considering commercial ACC systems [9,10,15,21], we find that the calibrated model of the vehicle tested in this work is neither L 2 or L ∞ strict string stable. The results are consistent across the different calibration methods, as summarized in Table 4.…”

“…Recently, as the vehicle systems have transitioned from research to practical deployments on commercial vehicles, the traffic modeling community is now in need of good models for how these vehicles behave in practice. Surprisingly, all commercial systems that have been tested [9,15,18,19,21,33] are shown to be string unstable and with varying performance characteristics. Being able to characterize the behavior of the ACC system in real time has implications for traffic management, where an emerging area of research [6,9,38,41,44] aims to dampen phantom traffic jams caused by string unstable driving behavior [34].…”

“…Punzo and Simonelli [29] estimated optimal parameters for several different car-following models on a four vehicle platoon equipped with GPS tracking units, and used a batch optimization method in which a gradient-based optimization technique was used starting from several different initial points, again in order to account for potential non-convexity. This method (along with ones in [9,10,15,18,21]) is in the same family of techniques used in this article as a benchmark to compare the newly presented methods. In general it should be noted that many of these techniques find parameters of non-linear models, and as such need to employ optimization techniques that account for this difficulty.…”

“…then the model is said to be L ∞ strict string stable [25]. All studies that have collected empirical data on commercial ACC systems have found them to be string unstable [9,10,15,21] (in the L 2 sense). In Section 5, we illustrate that the new online methods introduced in this work find that a stock 2019 SUV is neither L 2 nor L ∞ strict string stable.…”

Modeling the safety and efficiency performance of automated vehicles using fuzzy logic

“…The instability can be solved through V2V communication, which is the key to have the highest possible gain from AVs as stated by Shladover (57). Stability is also studied in the work of Knoop et al (59), which describes an experiment with seven SAE level-2 vehicles driven as a platoon for a trip of almost 500 km. The traffic dynamics in this experiment showed that the platoon becomes unstable when all vehicles have ACC activated.…”