1. Introduction

There are several reasons why modelling is important and also a challenge in the light of Automated Vehicles. Modelling is important:
  • Estimate the impact of different levels of automation on traffic flows and road capacities
  • Help train and develop the software that controls and monitors the performance of automated vehicles
  • To estimate future impacts of AVs in order to develop better transport plans and a more realistic estimation of costs and benefits for projects that will be operational when autonomous vehicles will be on the road.

Each of these objectives will require the use a different set of tools.

2. Modelling

There are a range of existing, longstanding modelling and forecasting tools and techniques available to support the assessment of vehicle safety, network operation and longer-term planning aspects and travel patterns in relation to conventional vehicles. However, for the assessment of CAVs many of these tools and techniques require review and in all likelihood some degree of adaptation. Furthermore, in many instances there a need to develop entirely new and often more complex tools and techniques to respond to the many and various new scenarios CAV technologies present. Correspondingly, there is the need for previously either unrelated or loosely related industries and disciplines work much more closely. Some examples of these are:
  • automotive engineers/modellers
  • highway and ITS designers/engineers
  • land use planners/modellers
  • strategic highway modellers
  • strategic multimodal and demand forecast modellers
  • telecommunications engineers
  • traffic engineers/modellers
  • transport demand modellers/forecasters
  • urban realm designers/architects

Up until the relatively recent past the underlying behaviour relating to private vehicle trips has remained largely the same. For a typical trip, the driver is usually the vehicle owner or co-owner, chooses to make a trip, when to make it, the route to take and any changes to the chosen route along the way in response to network conditions. Furthermore, throughout the journey and on a second-by-second basis, determines how the vehicle interacts with other road-users and external controls/advice such as speed limit signs, traffic signals and variable message signs. In short, though much has changed in terms of vehicle technology in areas such as ride, comfort, performance, fuel efficiency etc... the other aforementioned aspects haven't changed a great deal inasmuch as both driver and vehicle free from external influences or override. So in the context of CAVs, with regards 'C' - vehicles have been unconnected (with the exception perhaps for radio broadcasts on traffic conditions), and with regards 'A' - the autonomy is with the human driver. Considering now modelling tools and techniques, given the relatively unchanged environment described, the fundamental approaches for modelling existing and then inferred future conditions in a given road network have remained fairly consistent. This applies both to localised/detailed modelling (used in traffic engineering) and strategic modelling looking at wider network effects and travel routing/re-routing.

However, as an example of a behavioural change requiring some review and adaptation of modelling assumptions - over the past decade or so there has been a steady growth in the use of in-car navigation (e.g. Waze, TomTom, Google Maps, Apple Maps etc..), whether hand-held/smartphone or integrated in-car systems (in England, UK this had reached just over half the vehicle fleet by 2015). With this, an early example of 'connectedness', vehicles receive route advice, information on network conditions and accordingly, revised route advice and/or estimated time to destination. Whilst this change does not fundamentally change modelling assumptions (vehicles will still be trying to find their least cost/most efficient way through the network to their destination) there will be some behaviour change as compared to if all vehicles were 'unconnected'.

3. Model types

Model used to assess
Open source examples
Proprietary examples
Land Use, Balance between transport modes (multimodal) e.g effect of ticket prices, fuel prices, and other policies etc.., National, Regional or city-wide effects e.g. Highway Assignment Model (HAM) – effect of traffic growth and/or new developments, road widening schemes etc…. base and future year forecasts of traffic (re)assignment, delays. Macroscopic models generally assume steady state conditions over a modelling period and therefore trips and tours remain uniform and constant during one hour or more.
Generally aggregate but disaggregate models are also used. Models can be trip, tour or activity based.
Matsim, but rarely used in practical applications
CUBE, Visum, EMME, TransCad, Ainsum, Saturn
Mesoscopic models have been developed to capture the dynamics of trip making and congestion with delays affected by explicit queues and absolute capacity constraints. Some model of queueing, therefore, is needed and these models sometimes use a mix of Aggregate/Agent based in the same model, typically to speed up/lower computational load for parts large agent-based models.
Aggregate/agent based mix
Saturn, Avenue, Visum, Vissim, Aimsun
Detailed performance of highway networks from one junction, a local high street up to a small town around 4-5 sq miles typically (urban), larger if motorway/corridor (vehicular microsim).

Similar for pedestrian microscopic modelling. From a train platform, up to a station, up to a large area/event Olympic Park, Music/Festival event.

Plenty of scope for further work at this level with regards C/AVs, MaaS and
Matsim, Sumo, JuPedSim
Aimsun, Vissim, Paramics, Legion, MassMotion, VisWalk, STEPS
Automotive engineering/design/safety testing etc… High-fidelity modelling.
Gazebo, ROS
Pre-scan, Carmaker, Matlab
Note: there are quite a few examples of interfaces between different levels of models and many examples were the data inputs are similar or the same. Equally, it is common for model iterations to feed up and down. e.g. HAMs producing future year traffic patterns which are fed into microscopic models, junctions, links and signal timings revised and then the HAM revised and re-run.

Particularly in respect of CAVs, there are a number of examples of API links between Nano and Microscopic models so there are one or a number of ego CAVs (nano) driving and interacting with human-driven vehicles (and vice versa) in the micro network, so-called co-simulation or meta simulations.

MaaS is also creating the need for agent-based models at a Macroscopic level and TSC has been developing workflows/modelling in this area, though it is a ‘long game’ in terms of developing a robust approach due to scale and complexity involved.

4. Modelling CAV performance

Most studies of CAV traffic flow performance have been undertaken using micro-simulation and assumptions about CAVs characteristics contrasted with that of human-driven vehicles. Adherence to speed limits and legal manoeuvres, minimum headways and desired acceleration and deceleration rates have been assumed to represent different levels of automation and the impact of a mix of them on different types of roads. For the UK, the Department of Transport commisioned a study to do precisely this. The work was undertaken by Atkins and there is a Report on CAVs available. There are several other examples in different countries of this report summarises some of them.

To an extent, this is relatively straightforward. However, CAVs will not only affect traffic flow. As stated elsewhere in this site, CAVs may also induce traffic with new users, additional mileage and some empty travelling to serve other users or park, may take demand from public transport and release space previously used for parking. Modelling this type of movements is more complex and with many uncertainties. The choice of using CAVs as MaaS or owning them is critical in this case as the movements generated in each case will be very different. Software developers are trying to grapple with these issues and some studies have attempted to simulate what would be effects on traffic, congestion and emissions under different assumptions.

Of particular interest is the extent of adoption of CAVs as MaaS. Depending on the assumptions adopted it is possible to show some very dramatic impacts. The work undertaken by the International Transport Forum (ITF) is interesting as it compares the simulated impact of different levels of MaaS in different cities (Lisbon, Dublin, Auckland). These studies do not require MaaS to be provided by automated vehicles but the implications for scheduling services are very similar. ITF notes that MaaS is particularly attractive when all car, taxi and bus trips are replaced with shared mobility services whilst all rail-based modes are retained. Shared mobility services are scheduled to serve all trips with either shared taxi-like services or flexible route mini-buses. The waiting time and diversions needed to serve all trips are thus minimised. In the case of Helsinki, under these conditions, a 23% in Vehicle Kilometres travelled can be achieved. The problem is, of course, that it is generally considered unfeasible to force this level of reduction in car trips; it must be achieved gradually and through appropriate transport policies.

5. Modelling the impact of CAVs on large projects

The problem of accounting for CAVs when planning cities or the feasibility of major infrastructure or public transport project is a significant one. These large projects have a long lifespan and will operate simultaneously with CAVs. Given the uncertainties involved it seems difficult to be certain about the balance between owned and hired ( MaaS) automated vehicles and the impacts of them are different. It would seem appropriate to attempt to model a range of scenarios with rates of penetration of both CAVs and AV MaaS.

As a minimum, it would be appropriate to adopt some estimate of the impacts on road capacities and willingness to pay to save time, as recorded by our Expert Panel. A further step would be to make assumptions about the balance of owned and hired CAVs with their impact on trip induction. Other impacts, for example on parking, are more difficult to model at this stage, at least for project appraisal objectives.

It would be very useful to collect data on the experience of what may be called precursors to AVs as MaaS: taxis, shared mobility services, flexible mini-buses where available. This will provide a better indication of the rate of adoption of such services than simulations with assumptions about this factor.

6. Detailed Automotive and Network Modelling for CAVs

Given the safety-critical nature of vehicle design and development, the virtual testing software is similarly detailed. Vehicle systems, electrics, mechanics, suspension and sensors analysed to a very high level of fidelity (microseconds).

CAVs provide a range of new technical equipment and associated complexity which needs to be understood. In a modelling sense, a vehicle's eyes, ears, driver decision making and actions are transferring from the human driver over to the vehicle. Accordingly, safety assessments need to robustly capture and understand the likely behaviour of the vehicle in a wide range of situations.