Demand estimation and fleet management for autonomous mobility on demand systems

Mobility On Demand (MOD) systems are creating a paradigm shift in transportation, where mobility is provided not through personally owned vehicles but rather through a fleet of shared vehicles. To maintain a high customer quality of service (QoS), MOD systems need to manage the distribution of vehicles under spatial and temporal fluctuations in customer demand. A challenge for MOD systems is developing and informing a customer demand model. A new proactive demand model is presented which correlates real-time traffic data to predict customer demand on short timescales. Traditional traffic data collection approaches use pervasive fixed sensors which are costly for system-wide coverage. To address this, new frameworks are presented for measuring real-time traffic data using MOD vehicles as mobile sensors. The frameworks are evaluated using hardware and simulation implementations of a real-world MOD system developed for MIT campus. First, a mobile sensing framework is introduced that uses camera and Lidar sensors onboard MOD shuttles to observe system-wide traffic. Through a principled approach for decoupling dependencies between observation data and vehicle motion, the framework provides traffic rate estimates comparable to those of costly fixed sensors. Second, an active sensing framework is introduced which quantifies demand uncertainty with a Bayesian model and routes mobile sensors to reduce parameter uncertainty. The active sensing framework reduces error in demand estimates over both short and long timescales when compared to baseline approaches. Given estimates of customer demand, the challenge for MOD systems is maintaining high customer QoS through fleet management. New automated fleet management planners are introduced for improving customer QoS in ride hailing, ride requesting, and ridesharing MOD operating frameworks. The planners are evaluated using data-driven simulation of the MIT MOD system. For ride hailing, to address the challenge of missed customers, a chance-constrained planner is introduced for positioning vehicles at likely customer hailing locations. The chance-constrained planner provides a significant improvement in the number of served hailing customers over a baseline exploration approach. For ride requesting, to address the challenge of high customer wait times, a predictive positioning planner is introduced to position vehicles at key locations in the MOD system based on customer demand. The predictive positioning planner provides a reduction in service times for requesting customers compared to a baseline waiting approach. For ridesharing, incorrect assumptions on customer preference for transit delays can lead to poor realized customer QoS. A ridesharing planner is introduced for assigning customers to vehicles based on a trained ratings-based QoS model. The ridesharing planner provides robust performance over a range of unknown customer preferences compared to approaches with assumed customer preferences.

Department dc:contributor.department

Massachusetts Institute of Technology. Department of Mechanical Engineering.

Grantor dc:publisher

Massachusetts Institute of Technology

Year dc:date.issued

2017

Author dc:creator

• Miller, Justin (Justin Lee)

Advisor dc:contributor.advisor

• Jonathan P. How.

• Mechanical Engineering.

Statement dc:rights

• MIT theses are protected by copyright. They may be viewed, downloaded, or printed from this source but further reproduction or distribution in any format is prohibited without written permission.

Licence dc:rights.uri

• http://dspace.mit.edu/handle/1721.1/7582

Language dc:language.iso

eng

Handle dc:identifier.uri

http://hdl.handle.net/1721.1/113541

OAI identifier oai:identifier

oai:dspace.mit.edu:1721.1/113541