Network performance of autonomous cars at low market shares

Abstract

In this paper we consider how network performance is affected by a mixture of two heterogeneous car classes, a class which is designed to mimic the behaviour of autonomous vehicles (AVs) and a class which is designed to mimic normal driver behaviour. This makes it possible to investigate network effects as a function of the market shares of AVs. It is relevant from a policy and planning perspective as it will support decision making in the long-sighted phasing-in period of AVs.

In the paper, we base the AV class on a non-stochastic formulation of the intelligent driver model (IDM) due to Treiber et al. (2000). This model represents an advanced adaptive cruise control (ADAS) system which is “collision-free” and likely to find its way into autonomous cars. The segment representing normal driving behaviour is based on a modified and stochastic representation of the IDM model. Hence, in the latter model we allow for heterogeneity among drivers and introduce changes to the parametrisation to better reflect normal driver behaviour. We base the parametrisation of the randomness on experimental measurements from the literature.

The hypothesis put forward in the paper is that even moderate to low market shares of AVs could have significant impact on the network performance. Hence, it is possible that AVs can act as “facilitators of smoothness” in the transport system. In congested networks where the car density is high the properties of the AVs will translate to other cars simply as a result of natural v2v dynamics. In a way these cars act in a similar way as optimal speed-postings implemented on many motorways around the world and with a proven record for improving the network throughput. In Copenhagen it was implemented for the M3 motorway and in a before and after study it was evidenced that speed-postings reduced the travel time in the magnitude of 10%.