Autonomous vehicles: Early applications Revision as of 17:52, 12 December 2017 by JulieDunbar
This page is open for editing because it is part of the Incubator. Have something to add? Please register so you can contribute. Have an option you would like to share? Please click on the 'Talk' button to enter the dialogue. The TF Resource Volunteers appreciate your feedback and interest.
Several agencies have already begun to integrate connected and autonomous vehicles into their travel forecasting process. All are "works in progress," such that the descriptions below are likely to evolve over the next year in response to changing analytical needs, knowledge is gained about social and behavioral responses to CAVs, and best practices emerge. Some examples of these early applications are described below, organized by broad category of models.
NCHRP Report 20-102(9) Providing Support to the Introduction of CV/AV Impacts into Regional Transportation Planning and Modeling Tools is work in progress to investigate the methods and process for considering CAV in travel modeling. One initial product from this work is a review of current practice, documented in Technical Memorandum 1: Review of Recent AV CV Modeling, December 2016. The following material comes from this report and is intended to demonstrate existing work/experimentation for evaluating the impacts of CAV using travel forecasting models and tools. This is by no means an exhaustive list of existing work and others who have conducted work of a similar nature are encouraged to include a description of the approach, measures and outcomes here.(link to blank page on Other CAV Existing Work)
It is important to note that none of the modeling frameworks described here are necessarily better than the other when it comes to accommodating the uncertain future of CAV. Whether it is the overall lack of behavioral data that exists when considering CAV impacts or the caveat of imposed changes intended to represent expected behavioral changes from widespread deployment of CAV, there remains much to be learned. However, some experimentation has been done and is reported here to begin to provide some understanding of the complexities and challenges to be faced.
These examples are divided into 3 categories. Those using : • Trip Based Models, • Activity Based Models, and • Other Modeling Frameworks.
Trip-based modeling systems
Each of the four steps in a trip-based model (trip generation, trip distribution, mode choice and traffic assignment) can be modified to include some aspect of CAV technology. Potentially modified parameters or processes in the four-step model stream are:
- Regional geographic distribution of household and employment growth inputs.
•Value of time in generalized cost equations and mode choice. •Modifying in-vehicle travel time and other mode choice parameters. •Adding a mode for AV and estimating associated parameters. •Post processing of trip tables for input to disaggregate traffic assignment. •Modifying network link capacities. •Re-designating trips from SOV to a new CAV mode, or from SOV to high occupancy vehicle (HOV) modes to reflect how CAV might impact ridesharing. •Modification of trip rates (person or auto, truck, commercial)
Techniques used for CAV modeling reported are focused on changing trip tables to mimic potential changes in behavior, modifying mode choice parameters, and changing network capacity to reflect potential operational characteristics of CAV.
Capital Area MPO (Austin, Texas) CAV Modeling Experiment
Activity-based modeling systems
Tour and activity-based models are typically implemented in a microsimulation framework, making addition of ad hoc components and capabilities easier than in the aggregate deterministic frameworks employed in trip-based models.
San Francisco Bay Area
The ability to explicitly represent CAV demand and impacts was recently added to Ontario's provincial model. The modeling system, known as the Transport and Regional Economic Simulation of Ontario (TRESO), is a microsimulation-based modeling system that integrates local and long-distance resident, visitors, and commercial vehicle travel models with a space-time traffic assignment operating at two levels of network resolution. The specific enhancements relevant to modeling connected and autonomous vehicles include:
- Vehicles are added to the synthetic population (household and persons) based upon user-specified rules of CAV5 adoption by market segment. The possibilities include conventional and autonomous vehicles by SAE automation level that are either privately owned or shared. The latter are intentionally vaguely defined to enable travelers to choose the service or mode with highest utility during mode choice. The markets can be segmented by income, household structure, area type, or other household or person attributes.
- CAVs and mobility services (e.g., Lyft, Uber) have been added to the mode choice model, both as top-level choices as well as access and egress modes for various transit submodes.
- A ride-pairing module has been added to match inter-household trips by user-defined criteria and market shares to reflect potentially increased ride-sharing in both CAV and traditional vehicles. The matches are often made for similar travelers moving between the same origin and destination within a given time slice, which can vary by origin and/or destination region or accessibility levels at the origin or destination.
- The capacities in traffic assignment are adjusted based upon the degree of market penetration by CAVs implied during vehicle synthesis. The adjustments are based upon a methodology advanced by Levin & Boyles (2015) and traffic flow and vehicular considerations shared by Mahmassani (2016).
The enhancements intentionally require the analyst to explicit code assumptions about the future instead of relying upon estimated or asserted model parameters. Thus, the capabilities sacrifice rigor for ability to specify a wide range of alternatives. The intention is to run the TRESO system with a bundle of assumptions, enabling scores of different combinations of assumptions to be compared through the mining of model outputs.