Modeling impacts of autonomous vehicles on travel demand with Activity-Based Model Peter Vovsha, Jim Hicks, Ashish Kulshreshta, Surabhi Gupta (WSP) Vladimir Livshits, Kyunghwi Jeon (MAG) Session “Automated/connected vehicles” May 15, 2017

They are coming! Travel demand Vehicle routing, empty trips Network performance Improved mobility & accessibility Vehicle sharing, MAAS Connectivity, optimization of road & intersection capacity

Modeling aspects & assumptions in previous attempts to model AVs Direct impacts: Secondary assumptions: Increase auto availability Decrease auto time coefficient due to convenience & productivity Decrease auto access time and parking search time Increase freeway capacity Increase auto occupancy Increase trip rates for non-work and non-home-based trips Decrease parking cost Modal shifts Less demand for parking More demand for pocket lanes and pick-up/drop-off facilities

3 possible approaches Adapt existing travel model: If not an advanced one it will require many a priori external assumptions Develop a stand-alone QRS: Can serve long-term strategic purposes More flexible but hardly a replacement for detailed modeling Adjust advanced ABM: Back to behavioral foundations Revise network procedures

MAG CT-RAMP2 Approach 100% penetration of AVs: AVs owned: Partial penetration version is on the way AVs owned: Shared AVs will require and additional layer Focus on primary impacts: Try to avoid secondary assumptions but some are necessary

How will autonomous vehicles affect travel demand and traffic? (1) Elderly, youth, disable, and other people w/o driver license will have access to cars ABM: SOV, HOV/driver modes not constrained by age

How will autonomous vehicles affect travel demand and traffic? (2) Cars available at any location any time, not necessarily from home for entire tour ABM: Trip mode combinations on the tour less restrictive with any sequence of auto and transit Home Work Shop Shop

How will autonomous vehicles affect travel demand and traffic? (3) Empty repositioning trips made by AVs ABM: Certain travel tours with 1 destination and long duration (4h+) may have car repositioning trips to and from home Home Work

How will autonomous vehicles affect travel demand and traffic? (4) General convenience of AV-KNR versus PNR and transit with walk access/egress: ABM: KNR convenience parameters equalized to auto Home KNR Work

How will autonomous vehicles affect travel demand and traffic? (5) In-Vehicle Time Productivity: ABM: productivity “Bonus” for premium transit (-25% of IVT) applied to CAV

How will autonomous vehicles affect travel demand and traffic? (6) Optimized use of highway capacity, more efficient driving, increased intersection capacity: ABM: Assumed link capacity growth 5-10%

Trips by Mode: CAV vs. Baseline 2035 Substantial growth in auto “driver” trips Elimination of many passenger, walk, taxi, and school bus trips Shift of transit users to KNR

Trips by Mode (Impact on Transit): CAV vs. Baseline 2035 Shift of transit users to Kiss and Ride

Le Vine et al, 2017 Scenarios based on assumptions on upper limit of vehicle deceleration and reaction time

AVs freeway capacity improvements (Mahmassani, 2016) Leader Follower Spacing Connected Sc (73 feet) Autonomous Sa (73 feet) Regular Sr (146 feet)

AVs freeway capacity improvements (Mahmassani, 2016) Event Probability Spacing Connected vehicle (CV) Pc Autonomous vehicle (AV) Pa Regular vehicle (RV) Pr CV follows CV (Pc)2 Sc (73 feet) AV follows any vehicle Sa (73 feet) RV follows any vehicle or CV follows AV 1- Pa - (Pc)2 Sr (146 feet) Mix Sc(Pc)2 + SaPa + Sr[1-Pa-(Pc)2 ]

AVs freeway capacity improvements (Mahmassani, 2016) Regular vehicles Connected vehicles Autonomous vehicles Average lane capacity, veh/h 100% 0% 1,800 50% 2,057 2,400 2,880 3,600

Conclusions on current State of Practice in modeling AVs A lot of uncertainty and need for thinking out of box, especially for MAAS Need to focus on behavioral foundations and modeling of direct impacts Modeling secondary assumptions is less useful and can be misleading Advanced ABM offers useful features for modeling AVs Gradual AV penetration scenarios have to be considered

Contact(s) Peter Vovsha, PhD Vladimir Livshits, PhD Assistant Vice President, WSP Systems Analysis Group Peter.Vovsha@wsp.com Vladimir Livshits, PhD Maricopa Association of Governments (MAG) Program Manager VLivshits@azmag.gov