AA4CC team experimented with several commercial cars with adaptive cruise control systems and demonstrated string instability

Friday, December 6, 2019, International Public Airport Mnichovo Hradiště: the team from Faculty of Electrical Engineering Czech Technical University lead by professor Zdeněk Hurák conducted a unique experiment with commercial cars equipped with adaptive cruise control (ACC). Owners of such cars have been found through a public call and they participated in the experiment at their own expenses and risks, hence an excellent instance of citizen science.

All the cars with their ACC systems switched on were driving on a 2-km long taxiway back and forth for a few hours, following a leader vehicle, thus forming a platoon. After the velocities of vehicles (nearly) settled, the leader adjusted its velocity mildly and the responses of the ACC-equipped vehicles were recorded using GNSS receivers and accelerometers.

In particular, in one scenario the leader would decelerate from 60 to 50 km/h and after some 5 s would accelerate back to the original 60 km/h.

Results of the experiment

Drivers in the vehicles towards the tail of the platoon could observe that in response to the velocity dip of the leading vehicle, the speed of their own vehicle would go as low as 30 km/h and then soar to some 80 km/h, which was initially set as their cruising speed (20 km/h above the speed of the leader), after which they would finally settle to the 60 km/h of the leader. A snapshot of the experimental data is attached below the text. The observed phenomenon is known as string instability. It is well known that this phenomenon is induced by human drivers and that it is the reason for the no-reason traffic jams. It was hoped that adaptive cruise control systems, once present in a large(r) proportion of cars on highways, would bring relief from such traffic jams. The experiment confirmed the worries of some that such hopes would come in vain.  

The data from the experiments can be found at https://gitlab.fel.cvut.cz/aa4cc/acc/lkmh. Some technical report(s) will be added later.

Media coverage

Zdeněk Hurák interviewed by Ondřej Topinka in Studio 6 ČT on Friday, December 6, at 6:52. [Link]

Report by Martin Tyburec of Česká televize from the event which made it to the prime time  evening news on Friday, 6.12.2019. [Link] (contains a link to a video).

Report by Vojtěch Koval of Český rozhlas Radiožurnál from the event. [Link] (contains links to both audio and video).

Article by Radek Pecák "Může adaptivní tempomat zabránit zácpám? Vědci z ČVUT to ověřili při experimentu" on Automix.denik.cz website, [Link].

What was not analyzed

In the experiment the commercial ACC systems were not compared against human drivers. It was not analyzed if ACC systems contribute to string instability less than human drivers do or if they make the situation even worse. On one hand, since ACC are machines, they are reliable, accurate and fast. On the other hand, they lack the human driver capability to anticipate, be it through seeing the red braking warning lights of the predecessor vehicle(s), and analyzing the traffic situation (work on the road, ending lane, ...).     

In the experiment, individual ACC systems (cars) were not compared against each other. Due to the format of the experiment, time was too short to conduct experiments with individual cars and it is only through such detailed experiments that a contribution of each vehicle to string instability could be quantified. It is well possible that each car make and model could be different in this regard.

State of the art

This study was not the first to demonstrate string instability with commercial ACC systems. In May 2019, a draft of a paper by Gunter et al. was made available, in which the authors document a similar experiment (seven vehicles with ACC and one leader). Even earlier (in 2014) a paper by Milanés and Shladover reads "This experimental result shows for the first time that the theoretical results previously presented in the literature [...], where car-following policies based only on predecessor information were used for ACC controllers, are inherently unstable when applied to real vehicles."  

In this regards, the present experiment could be viewed as yet another experimental confirmation that the currently deployed commercial ACC systems do not achieve string stability.


Things could change in future because in principle (according to the theory), string stability could be achieved within the ACC framework, but it may be the case that this requirement is orthogonal to the requirement of individual vehicle comfort, which is currently emphasized by car makers. 

But it is also possible that several realistic constraints such as delays induced by processing the data from the sensors (radar and cameras) make the ideal (predicted by the theory) performance unachievable. In this case, an improvement in performance of these advanced driver assistance systems can only be expected after incorporating other functionalities such as reading the visual information about braking of the predecessor vehicle encoded in the the red braking warning lights, or even encoded in a wireless vehicle-to-vehicle (V2V) communication. These are subject of intensive research in the engineering community in the last decade or so (Cooperative Adaptive Cruise Control, CACC) and several successful experimental demonstrations have been reported. An obvious disadvantage is that such systems will only be useful once a significant proportion of cars on the roads are equipped with them.

Image: Snapshot of experimental data from the experiment with ACC