#329 Development of Safe, Profitable, and Fair Robotaxi Deployment Strategy

Principal Investigator
Ding Zhao
Start Date
July 1, 2020
End Date
June 30, 2021
Research Type
Grant Type
Grant Program
FAST Act - Mobility National (2016 - 2022)
Grant Cycle
2020 Mobility21 UTC


The project aims to develop a safe, profitable, and fair deployment strategy for robotaxis by studying the possibility to deploy a fleet of autonomous vehicles with different functionalities. The requirement for AVs will be studied using the traffic primitive method and synthesized with transportation demands. Strategies will be developed to minimize the costs by commanding AVs with different functionalizes to appropriate routes while maintaining a similar level of safety standard. Deployment cost and average waiting time in different communities will be studied to balance the business cost and social benefits.    
Motivation and Goal of the Project

2020 is a critical year for the autonomous driving industry. According to Gartner inc, the Level 5 autonomous driving technology is passing the “peak of inflated expectations” and Level 4 has entered the zone of “trough of disillusionment”. Investors start to worry about the soaring cost to deploy autonomous vehicles (AVs) while the public is concerned about the safety and fairness issues potentially brought by the new mobility. As leading companies such as Waymo, Uber, and AutoX have obtained permits to commercialize their robotaxi fleets domestically and overseas, it is urgent to investigate deployment strategies that are sustainable for companies to run its business and beneficial to the public, so that the development of autonomous driving technology would keep its momentum and move to “slope of enlightenment” - the next phase in the Gartner hype cycle for emerging technologies after “trough of disillusionment”.

In this project, we aim to develop safe, profitable, and socially fair robotaxi deployment strategies. We found the related research is surprisingly sparse. Most research focuses on one aspect of the problem with ideal assumptions for other aspects. Assumptions could be 100% penetration of AVs, perfect automation technologies, almost unlimited budgets, etc. As the AV technology goes beyond the peak of inflated expectations, it is critical to developing technologies that are as practical as possible, which, instead of developing methods under idea assumption, seeks tradeoff between the metrics. Particularly, we plan to investigate the issue of how to trade-off between safety, profitability, and fairness when deploying AVs.

Research Methodologies

We will develop this project based on our previous projects. In PI Ding Zhao’s previous projects supported by Traffic 21, Denso, and Toyota, real-world driving data has been collected in Pittsburgh and driving risk has been analyzed geographically using Bayesian nonparametric learning. We created the accelerated evaluation method [1,2] to evaluate driving safety. We further used Nonparametric Bayesian learning to process the logged driving data and identify typical driving scenarios [3] The functionalities of automation are thus related to road and traffic condition [4]. In Co-PI Fei Fang’s previous work supported by Mobility 21 and Lawrence County, efficiency and fairness of ride-sharing have been modeled and analyzed using game theories. It was shown that fairness can be improved without intruding efficiency [5]. Thanks to the platform of the Smart Mobility Connection series seminars, the two PIs start to understand each other’s work and realize the two studies are complementary. By combining the vehicle level safety analysis and city-level efficiency and fairness analysis, it is possible to propose a practical deployment strategy improving safety, cost, and social fairness at the same time.

One key idea to achieve the goal is to promote a fleet of AVs with different functions. For instance, vehicles being driven on the highway may not need sensors and processors to do parallel parking; vehicles being driven close to a school zone does not need sensors to scan objects a few hundred meters away and update the object detection as frequent as on the highway as the velocity is small. Some cars may only drive in the daytime while others may be designed to drive in the night. The research team will synthesize the dynamic map of traffic demands with the driving functionality map to develop a safe, efficient, and fair deployment strategy. Equilibriums of the three aspects will be analyzed on high dimensional manifolds to have a good balance. We will start with a determinant setting and seek an adaptive approach to solve the problems. We will then consider the uncertainty in the demand and driving conditions and estimate the mean and variance of the proposed algorithm. 

[1] D. Zhao, “Accelerated Evaluation of Automated Vehicles,” Ph.D. dissertation, University of Michigan, Ann Arbor, 2016.
[2] A. Mansur, P. Glynn, and D. Zhao. "An Accelerated Approach to Safely and Efficiently Test Pre-produced Autonomous Vehicles on Public Streets." IEEE Intelligent Transportation System Conference, 2018.
[3] W. Wang and D. Zhao, "Extracting Traffic Primitives Directly From Naturalistically Logged Data for Self-Driving Applications," in IEEE Robotics and Automation Letters, vol. 3, no. 2, pp. 1223-1229, April 2018.
[4] Z. Jiacheng, W. Wang, and D. Zhao. "A Tempt to Unify Heterogeneous Driving Databases using Traffic Primitives," IEEE Intelligent Transportation System Conference, 2018.
[5] Ma, Hongyao, Fei Fang, and David C. Parkes. "Spatio-temporal pricing for ridesharing platforms." arXiv preprint arXiv:1801.04015 (2018).    
In order to achieve the goal, the research can be divided into the following tasks: 
Task 1: 6/1/2020-9/31/2020
Based on previous projects, synthesize the risk maps and traffic demanding models.  Estimate the cost of different functionalities of AVs. 
Task 2: 2/1/2020-6/30/2020
Develop deployment strategies for AVs with different functionalities
Task 3: 7/1/2020-10/31/2020
Evaluate the proposed methods with high-performance computing simulations
Task 4: 7/1/2020-12/31/2020
Implement the algorithms with the simulation of Pittsburgh. Propose potential suggestions to the city on transportation infrastructure for a smart city.
Deployment Plan
Task 1: Synthesize the dynamic traffic demanding model with the map of driving functionalities

In this task, our team will develop a dynamic approach to synthesize the dynamic traffic demanding model with the map of driving functionalities. The driving functionalities were developed in our previous projects based on a concept called traffic primitives, which describes the fundamental building block of the driving environment in a stochastic approach. We plan to extend the usage of traffic primitives from the vehicle level to the city level. To achieve this, we need to analyze the relationship between the traffic primitive and transportation infrastructure and analyze the relationship of primitives in geographically related zones. We plan to develop a novel representation for such information by using graph theories. The proposed methods will be able to simulate both the macroscopic transportation demand and the microscopic safety requirement.

Task 2: Develop deployment strategies for AVs with different functionalities

In this task, our team will design a deployment strategy to minimize the deployment cost while maintaining safety and fairness. The team will definite the “redundancy” of sensing and use it as an indicator to balance safety and functionalities. Currently, more than half of the cost of AVs are used to purchase sensing components. By reducing the unnecessary sensors for dedicated driving zones, the total deployment cost may dramatically drop. A mixed-integer optimization scheme will be proposed to study the effectiveness of the sensing components and command cars with different functionalities to different routes. The fairness will be investigated among districts with various infrastructure conditions. The study quantitively analyzes how the transportation infrastructure will affect the deployment cost of AVs thus provide unevenness for the communities living in different zones.

Task 3: Evaluate the proposed methods with high-performance computing simulations

We will evaluate the effectiveness of our methods with a comprehensive simulation. We plan to collaborate with the Lawrence Berkeley National Lab (LBNL). LBNL is developing a multi-model simulation tool for city-scale transportation for autonomous vehicles. The simulation can give a good performance but needs to be run on the high-performance computing facility at LBNL. We will work with colleagues at LBNL to remotely access the computing resource and integrate the safety analysis into the simulation scheme. The simulation algorithm now can simulate transportation in the Bay area. We plan to send one student to LBNL to be trained to operate the facility in the summer of 2020 and establish a simulation environment for Pittsburgh.

Task 4: Implement the algorithms with the simulation of Pittsburgh. Propose potential suggestions to the city on transportation infrastructure for a smart city.
We will apply our algorithms on the Pittsburgh transportation in simulation based on real-world data sets in the public domain and the TrafficNet database built in our previous projects and provide quantitative results. We plan to visit the city hall regularly to better facilitate the need of the Department of Innovation & Performance to deploy automated vehicles. We will also talk to our partners including Uber, Bosch, and Toyota so that they can benefit from this project thus being motivated to invest in this project and engage with the Mobility program. The team plans to open the methods, tools, and datasets to build a good ecosystem and thus flourish the automated driving community.

Expected Accomplishments and Metrics
Deliverables in this project include 
1. Novel methods to develop safe, profitable, and fair robotaxi deployment strategies.
2. Data sets, simulation results, and online/offline tools to deploy and validate the proposed methods.
3. A memorandum to the city of Pittsburgh to facilitate future policy-making.

The team will closely work with the Pittsburgh city and our industrial partners have granted $390,000 to support the research team to develop fundamental tools. Under the support of Mobility 21, we will be able to act as a bridge to link the needs of the industrial and the public.

Individuals Involved

Email Name Affiliation Role Position
feifang@cmu.edu Fang, Fei Carnegie Mellon School of Computer Science Co-PI Faculty - Untenured, Tenure Track
dingzhao@cmu.edu Zhao, Ding Carnegie Mellon School of Engineering PI Faculty - Untenured, Tenure Track


Amount of UTC Funds Awarded
Total Project Budget (from all funding sources)


Type Name Uploaded
Data Management Plan Mobility21_Deployment_Data_Management_Plan.docx March 11, 2020, 12:02 p.m.
Progress Report 329_Progress_Report_2020-09-30 Sept. 30, 2020, 11:20 a.m.
Progress Report 329_Progress_Report_2021-03-31 March 30, 2021, 8:58 p.m.
Final Report Final_Report_-_329.pdf Aug. 1, 2021, 9:51 a.m.
Publication Functional Optimal Transport: Mapping Estimation and Domain Adaptation for Functional data April 6, 2022, 5:13 a.m.

Match Sources

No match sources!


Name Type
City of Pittsburgh Deployment Partner Deployment Partner