Abstract
Connected autonomous vehicles (CAVs) are gradually advancing towards widespread deployments. CAVs promise to improve transportation safety by operating more efficiently and avoiding incidents like crashes due to human driver error. However, they may cause iincidents themselves, especially when interacting with humans. The goal of this project is to evaluate the potential safety benefits of CAVs in mixed-autonomy settings, in which CAVs and human vehicles share the road. Our work has three parts: (i) estimating the effective incident rates of CAVs and how they are distributed across a city, leading to algorithms for prioritizing incident responses so as to reduce their overall impact on traffic flow and safety; (ii) incorporating CAVs’ and human drivers’ ability to react to human pedestrians, leading to algorithms for CAVs to reduce pedestrians’ impact; and (iii) evaluating our models and analysis in our mixed-autonomy simulator.
Towards modeling CAVs’ effect on traffic incident rates, we will account for the fact that vehicle incident rates vary with the road congestion level and type, e.g., Pennsylvania data show that incidents are more common in heavy-traffic surface streets than sparsely populated highways. We will build on our prior Mobility21 work studying mixed-autonomy traffic patterns to account for changes in congestion levels across the road network due to vehicle incidents, e.g., if CAVs overall reduce the incident rate on highways, this might lead to better overall traffic flow and fewer subsequent incidents. Our results will enable prioritization of incident response so as to maximally reduce the resulting traffic congestion.
We will then incorporate the effects of human pedestrians into our mixed-autonomy setting. Pedestrians can change safety dynamics as their actions may be more difficult to predict, especially for CAVs that may not be well-trained on pedestrian data. For example, CAVs can improve traffic flow by more closely following other vehicles; this is less feasible when human pedestrians are present. We therefore plan to incorporate these pedestrian “shocks” into our model of traffic flow and incident rates. We will use these results to propose new techniques for CAVs to predict and plan for pedestrian behaviors.
We will use our existing mixed-autonomy simulator, developed with Mobility21’s support, to numerically evaluate our models and how the above safety effects vary for different amounts of CAVs. We will also leverage models and feedback from our deployment partner, the Southwestern Pennsylvania Commission (SPC), in our simulations. We will further measure how CAVs’ effects are distributed around a city and implications for equity (see also “Outputs” below).
This project is synergistic with our concurrently submitted proposal entitled “Mitigating Cascading Failures for Safety in Transportation Networks in the era of Autonomous Vehicles,” where the goal is to evaluate the safety impact of AVs from the perspective of their impact on cascading road failures and congestion. In contrast, the current project focuses on CAVs’ safety impact in terms of the traffic incident rate in mixed-autonomy settings. As such, the two projects complement each other and can be combined at a total budget of $150,000 if preferred.
Description
Timeline
Strategic Description / RD&T
This project addresses the impact of CAVs on safety, particularly considering the interactions of CAVs with human-driven vehicles and pedestrians. Thus, it addresses two priorities listed on page 16 of the US DOT RD&T Plan: “Vehicle and aircraft safety, automation, and connectivity” and “Vulnerable road user safety.” Our work will further help to achieve one of the desired outcomes listed on this page: “Emergency response times are shortened so that when crashes occur, victims receive effective treatment sooner and crashes are cleared more quickly to prevent secondary crashes.”
We believe that this project falls under the “Data-Driven System Safety'' research priority and objective. In particular, it will help to “[a]dvance transportation safety by evaluating the safety of existing transportation technologies and supporting the safe integration of emerging technologies.” (Table 3, page 17). Our work will numerically evaluate the safety of emerging CAV technology in realistic mixed-autonomy scenarios, thus supporting its safe integration into existing transportation systems. More specifically, the project’s focus on the effect of human pedestrians will “[e]valuate the effect of transportation technologies on safety outcomes for vulnerable populations'' (page 19), and its focus on CAVs will “[s]upport the development, evaluation, and implementation of connected digital infrastructure designed to enhance transportation safety outcomes” (page 19).
We finally note that on page 21, the RD&T Plan states that “Human error is to be expected, so road infrastructure and vehicle technology must be designed and operated so that deaths and serious injuries are managed through system safety engineering.” Our project similarly makes this assumption, and our models and simulator will incorporate errors made by human drivers and pedestrians in simulating their effects on incident rates.
Deployment Plan
We list our planned activities and deliverables for each quarter below.
Q1: Development of models for traffic incident rates with CAVs and algorithms to prioritize incident response. We will use our prior work on traffic flow and mixed-autonomy settings to model the effects of CAVs on overall incident rates, which can depend on congestion levels and road types. Our main deliverable will be new mathematical models of CAV movement and incident rates around a city.
Q2: Development of models that include human pedestrians and CAVs’ and human-driven vehicles’ reactions to unexpected actions by pedestrians. Our main deliverable will be to extend our mathematical models from Q1’s deliverable to include pedestrian actions.
Q3: Development of a mixed-autonomy simulator that quantifies incident rates on realistic road network topologies, using our models of CAV, human driver, and human pedestrian behavior. We plan to build on the simulator we had previously developed with Mobility21’s support. We will also consult with SPC on other potential sources of data and deployment and equity concerns that we should include in the simulator. Our main deliverable will be the simulator code and data, which we will release publicly.
Q4: Evaluation of our proposed models in the mixed-autonomy simulator. See below for specific anticipated outcomes and outputs of this evaluation. Our main deliverable will be a quantitative analysis of the impact of CAVs on safety.
Expected Outcomes/Impacts
By quantifying CAVs’ effects on traffic flow and accident rates with other vehicles and human pedestrians, we will enable regulatory agencies to anticipate the effects of gradual but increasing CAV deployment. Similar to our prior "Big Idea" project with Mobility21, on which this project builds, our work could inform the policies that agencies may wish to take in regulating CAVs (e.g., developing new metrics for how well they interact with human drivers and pedestrians). It will further provide quantitative validation for (or evidence against) claims that CAVs can benefit traffic and traffic safety as a whole. Finally, our simulator may aid such agencies and other researchers in studying CAV and human behavior in mixed-autonomy systems.
Emergency responders may benefit from our work by prioritizing responses to incidents that have the greatest overall impact (in terms of traffic congestion and follow-on incidents) on the road network.
CAV manufacturers may also be able to benefit from this work, as it will help them quantify the importance of developing AVs that interact “well” with humans and design learning algorithms that help CAVs to do so. These new learning techniques may enhance the benefits of CAVs and increase their potential for deployment.
Expected Outputs
Our concrete deliverables will be (i) algorithms for emergency responders to prioritize incident response according to overall impact on congestion and safety in the road network; (ii) new algorithms that allow CAVs to plan for (potentially unexpected) human pedestrian behavior; (iii) quantitative estimates of how much CAVs will improve traffic flow for different types of roads, as a function of their prevalence on roads; (iv) quantitative estimates of how CAVs affect incident rates between vehicles and pedestrians, for different CAV penetration rates; (v) regulatory implications of these quantitative findings; and (vi) extensions of our existing mixed autonomy simulator to include better models of human pedestrian and driver behavior.
We will work with our equity and deployment partner, the Southwestern Pennsylvania Commission (SPC), to ensure the usefulness of our project outputs, through regular meetings for feedback. We will also consider leveraging SPC’s existing public participation processes to receive community feedback on our work; analysis of this feedback will be another output of this work. We expect that this feedback will allow us to account for metrics that are important for all communities affected by CAV deployments, helping to ensure that we address equity concerns associated with CAV deployments.
TRID
We have attached the results of a TRID search for “mixed autonomy safety,” which returned 18 results. Seven of the returned projects focus specifically on intersection, lane, or ramp control; in contrast, our project will study the safety effects of traffic incidents and congestion throughout the entire road network, not just in specific intersections and ramps. Two of the returned projects focus on altruism and cooperation between CAVs and human-driven vehicles but do not consider the effects on safety. We plan to incorporate CAV cooperation in modeling CAV behavior in our existing mixed autonomy simulator. We will consider modifying our cooperation models according to these projects’ results.
We believe the most relevant results for our project are (i) a survey of state and motion prediction for vehicles and vulnerable road users (Result 4), (ii) a study of public acceptance of CAVs (Result 13), and (iii) a review of safety measures in CAV safety modeling (Result 14). We will consult these resources in defining our models of vulnerable road users (e.g., pedestrians) and quantitative metrics for safety in mixed-autonomy settings. We will use the results of the public acceptance to inform our work with SPC on equity and community considerations.
Individuals Involved
| Email |
Name |
Affiliation |
Role |
Position |
| cjoewong@andrew.cmu.edu |
Joe-Wong, Carlee |
ECE |
PI |
Faculty - Untenured, Tenure Track |
| ichengl@andrew.cmu.edu |
Lin, I-Cheng |
ECE |
Other |
Student - PhD |
| oyagan@andrew.cmu.edu |
Yagan, Osman |
ECE |
Co-PI |
Faculty - Research/Systems |
Budget
Amount of UTC Funds Awarded
$75000.00
Total Project Budget (from all funding sources)
$201765.00
Documents
Match Sources
No match sources!
Partners
| Name |
Type |
| Southwestern Pennsylvania Commission |
Deployment & Equity Partner Deployment & Equity Partner |
