Real-time adaptive signal control systems are limited by the accuracy of contemporary vehicle detection technologies. Regardless of the sensing modality chosen (video, thermal, radar, etc.), standard outputs consist of snapshots of vehicle counts and presence data as they pass through designated spatial regions on the roadway. The distance between such sensing zones typically results in considerable uncertainty (e.g., due to mid-block vehicle exits and entries from the roadway) in the adaptive signal control system’s predictive model of approaching traffic volumes and vehicle arrival times. Connected vehicle technology offers the promise of unprecedented improvement in predictive modeling accuracy through receipt of continuous location, heading and speed updates of approaching vehicles. Yet the full benefit will accrue only as the level of penetration of equipped vehicles increases, and full penetration is likely still decades away.
This project proposes to use the sensing abilities of connected autonomous vehicles (CAVs) to compensate for the relatively low numbers of equipped vehicles on the road in the shorter term, and accelerate the sensing benefit that can be provided to adaptive signal control systems by connected vehicle technology over time. The basic idea is to have the CAV communicate not only its continuous location, heading and speed to the infrastructure, but also the continuous location, heading and speed of the vehicles that it perceives in its local vicinity as it proceeds along the road, in essence “virtually” increasing the number of equipped vehicles on the road. To take advantage of the existing Pittsburgh connected vehicle testbed in East Liberty, we will focus specifically on establishing interoperability between CAVs and the Surtrac adaptive signal control system, and on evaluating benefit of this extended vehicle-to-infrastructure (V2I) communication in this context. DSRC protocols will be developed to enable CAV communication of information about surrounding vehicles, and the Surtrac system will be extended to integrate this continuously received location, speed, and heading information into its baseline predictive model (derived from its conventional vehicle detection devices). We will evaluate the performance benefit accrued from using this additional sensing information first in simulation, by analyzing improvement with respect to standard mobility metrics (delay, number of stops, wait time, etc.). Then, in collaboration with Argo AI and Rapid Flow Technologies, we will develop a field implementation and experimentally validate the approach within the East Liberty test bed.
Motivation and Overview
The emergence of connected vehicle technology offers tremendous opportunities for enhanced mobility in future urban and suburban areas through vehicle-to-infrastructure (V2I) communication. In the long term, when connected vehicle (CV) technology is ubiquitous across all travelers (passenger vehicles, freight carriers, transit, bicycles, pedestrians, etc.), it will offer unprecedented infrastructure sensing capability, to the extent that existing vehicle sensing technologies at signalized intersections (and perhaps even the intersections themselves) may become obsolete. However, practically speaking, this eventuality is still decades away. Without full penetration of connected vehicle technology, infrastructure systems that require real-time sensing of traffic on the road (such as adaptive traffic signal control systems) will have to continue to rely on data received from less accurate, contemporary vehicle and pedestrian detection technologies, with continuous V2I communicated location, heading and speed information being used to augment and improve the accuracy of this baseline data as equipped vehicles begin to appear on the roadways. The impact of this more accurate real-time information on traffic signal control performance will depend fundamentally on the number of equipped vehicles on the road and communicating their real-time information. A previous study has projected that the level of penetration at which V2I data can be expected to significantly upgrade traffic signal performance is between 25% - 40%. [Feng et. al. 2015]
We propose research aimed at accelerating the rate at which CV data can be made available to adaptive traffic signal control systems. Specifically, we propose to use the sensing abilities of connected autonomous vehicles (CAVs) to compensate for the relatively low numbers of equipped vehicles on the road in the shorter term, and accelerate the sensing benefit that can be provided to adaptive signal control systems by connected vehicle technology over time. The basic idea is to have the CAV communicate not only its continuous location, heading and speed to the infrastructure, but also the continuous location, heading and speed of the vehicles that it perceives in its local vicinity as it proceeds along the road, in essence “virtually” increasing the number of equipped vehicles on the road. Since it is reasonable to assume that CAVs will begin to become operational on roadways within the next five years or so, we believe this approach has the potential to substantially reduce the CV level of penetration barrier to more accurate sensing and predictive modeling.
To take advantage of the existing Pittsburgh connected vehicle testbed in East Liberty, we will focus specifically on establishing interoperability between CAVs and the Surtrac adaptive signal control system [Smith et. al. 2013], and on evaluating benefit of this extended V2I communication in this context. Work will be carried out in collaboration with Argo AI, a Pittsburgh-based self-driving car company, and Rapid Flow Technologies Inc, the commercial supplier of the Surtrac system. DSRC protocols will be developed to enable CAV communication of information about surrounding vehicles, and the Surtrac system will be extended to integrate this continuously received location, speed, and heading information into its baseline predictive model (derived from its conventional vehicle detection devices). We will evaluate the performance benefit accrued from using this additional sensing information first in simulation, by analyzing improvement with respect to standard mobility metrics (delay, number of stops, wait time, etc.). Finally, we will develop a field implementation and experimentally validate the approach within the East Liberty test bed.
Most research to date on using CV technology in conjunction with adaptive signal control to enhance mobility, including our own, has focused on mechanisms that provide explicit benefit to vehicles that are equipped and hence can be effective in settings where the level of penetration of equipped vehicles is low. In [Hawkes 2016], it is shown (in simulation) that if an equipped vehicle is willing to share its route with the infrastructure (in this case, the Surtrac adaptive signal control system), then it is possible for the control system to move this vehicle through its signalized network substantially faster than before (up to 20% faster on average) without adversely affecting vehicles that are not equipped - basically by providing the traffic control system more information, uncertainty is reduced and the system can do a better job of optimizing traffic flows. A second possibility, currently under investigation in collaboration with the Port Authority of Allegheny County, is to better predict bus arrival times at the intersection [Isukapati et. al.2018a, Isukapati et. al. 2018b], and to use this and additional real-time information provided by V2I communication (e.g., ahead or behind schedule, how full, route, door open and close events) to provide a smarter transit priority capability that takes all other traffic approaching or already waiting at the intersection into account. Finally, work is underway aimed developing a mobile app to allow pedestrians with disabilities to interact directly with the intersection and actively influence traffic control decisions to ensure safe crossing [Smith 2017]. In all of these applications, however, enhanced mobility follows from the expediting of CV equipped travelers. In contrast, [Liang et. al. 2018] considers the improved benefit to traffic signal control performance of using location, heading and speed information received from CAVs to improve the signal system’s predictive model (like is proposed here), and likewise the improvement that is possible if the signal system is able to control the speed of CAVs. However, this work is restricted to study of a single intersection and does not take advantage of CAV sensing capabilities.
The proposed research presents a number of technical tasks:
• Protocols for extended CAV V2I communication – One technical task involves the development of protocols for communication of the information that is sensed by the CAV concerning surrounding vehicles. Current V2I communication within the Pittsburgh CV testbed is accomplished via DSRC, and since the communication of sensed information about other vehicles is not anticipated in the current DSRC message standards, the technical question is how to best provide this extended information. One approach could be to have the CAV issue multiple Basic Safety Messages (BSMs), one for itself and one for each sensed vehicle in its surroundings. However, since the information is compiled from local sensing simultaneously, it is perhaps more efficient of more effective to send to the infrastructure as a unit (for example, is a message about a given vehicle more reliable if it is sent by itself or by another vehicle that has sensed its presence in the vicinity). We propose to analyze alternatives with respect to expected CAV sensing abilities and DSRC performance characteristics and develop an appropriate protocol for extended V2I communication. We will restrict attention to sensed vehicles in the vicinity of a CAV for initial empirical analysis, but will anticipate CAV’s ability to provide information on additional travelers as well (e.g., pedestrians, bicycles).
• Integrating received V2I vehicle data into the Surtrac predictive model – A second technical task is that of integrating received V2I BSM information (or consolidated BSM information that summarizes a set of sensed vehicles) into the predictive model that is produced by Surtrac using information from its current infrastructure sensors. Regardless of the sensing modality employed (video, thermal, radar, etc.), the standard outputs of these sensing devices consist of snapshots of vehicle counts and presence data as they pass through designated spatial regions on the roadway, and the distance between defined sensing zones typically gives rise to considerable uncertainty with respect to approaching traffic volumes and vehicle arrival times. From this snapshot data, Surtrac constructs a predictive model that consists of sequences of vehicle “clusters” that are either approaching the intersection (platoons) or are already at the intersection (queues) from different directions along with expected cluster arrival and departure times. Hence, the first step will be to develop an algorithm for mapping the BSM information received for a given vehicle to the corresponding vehicle in the coarser, baseline cluster model. Once a match is established, it may be necessary to propagate the consequences of this BSM to other clusters, following from the fact that inferences can be made about the timing of other vehicles that have been grouped into the same cluster or are situated in following clusters in the sequence. In some cases (e.g., when a bus is detected to be moving slower than what was predicted in the baseline cluster model), the extent of this propagation can be significant (e.g., a stopped bus will also stop traffic that is following it). Thus, the second step will be to develop this cluster adjustment procedure.
• Managing redundant communication – Since it is possible for a CAV to sense another CAV (or CV for that matter), the above model updating process must be sensitive to the possibility of receiving redundant information about a given vehicle. Minimally, this can result in the inefficiency of redundant updates, but if the reliability of the information is a function of its source, then there will be a potential consistency/accuracy issue if the issue of redundant information is ignored. We will develop a mechanism for detecting and appropriately reconciling multiple inconsistent messages concerning the same vehicle.
• Simulation analysis – To first evaluate the potential benefit of the approach, the extended Surtrac system will be configured to run with a VISSIM or Sumo microscopic simulation model of the Pittsburgh CV test bed road network (choice of which simulation system to use will depend on strength of facilities provided for simulating V2I communication). We will implement a procedure that simulates the sensing of other vehicles in close proximity by a CAV. Then we will experimentally assess the impact of V2I communication of CAV data on traffic signal control performance, both with and without the sensing of nearby vehicles turned on. We will also consider the percentage of CAVs in the network to additionally measure how CAV volume impacts signal control performance. We will measure traffic signal control performance using standard traffic flow efficiency metrics, including average delay, average number of stops, and average wait times.
• Field demonstration and evaluation – A final task will be to develop a field implementation of the approach, for demonstration and further analysis within the Pittsburgh Surtrac CV test bed. Using the developed V2I protocols, message construction and encoding software will be developed for transmission from the CAV’s DSRC On-Board Unit (OBU). Argo AI will assume responsibility for developing this component, including the procedures necessary to access their internal vehicle sensors for detecting and collecting information about nearby vehicles, and to turn this information into the appropriate message format for transmission. Similarly, message decoding and interpretation software will be developed to handle the interface to Surtrac on the DSRC Road Side Unit (RSU) side. Rapid Flow Technologies will assist in the development of this component. Once a field implementation is complete, we will conduct CAV run through experiments to calibrate the traffic flow efficiency of the network with and without V2I communication and sensing.
[Feng et. al 2015] Feng, Y., Head, K. L., Khoshmagham, S., & Zamanipour, M. (2015). A real-time adaptive signal control in a connected vehicle environment. TRANSPORTATION RESEARCH PART C-EMERGING TECHNOLOGIES, 55, 460-473.
[Hawkes 2016] “Traffic Control with Connected Vehicle Routes in SURTRAC”, MS Thesis,
CMU Robotics Institute, May 2016.
[Isukapati, et. al. 2018a] Isukapati, I., H. Rudova, G.J. Barlow and S.F. Smith, “Analysis of Trends in Bus Dwell Time Data for Real-Time Predictive Modeling”, Journal of the Transportation Research Board, To Appear, 2018.
[Isukapati, et. al. 2018b] Isakupati, I., C. Igoe, E. Bronson, and S.F. Smith “Generating Highly Predictive Probabilistic Models of Task Durations”, unpublished paper, November 2018 (Under first revision for IEEE Journal of ITS).
[Liang, et.al 2018] Liang, X., Guler, S.I. and Gayah, V.V. Signal timing optimization with Connected Vehicle technology: Platooning to improve computational efficiency. Transportation Research Record, accepted for publication.
[Smith et. al. 2013] Smith, S.F. G.J. Barlow, X-F Xie, and Z.B. Rubinstein, “ Smart Urban Signal Networks: Initial Application of the SURTRAC Adaptive Traffic Signal Control System”, Proceedings 23rd International Conference on Automated Planning and Scheduling, Rome, Italy, June 2013.
[Smith 2017] S. F. Smith, “Connecting Pedestrians with Disabilities to Adaptive Signal Control for Safe Intersection Crossing and Enhanced Mobility”, Research proposal submitted to DOT Accessible Transportation Technology Research Initiative, awarded July 2017.
We intend to carry out the proposed work over the one year period from July 1, 2019 to June 30, 2020. We anticipate progress to proceed according to the following timeline.
[ Months 1-3] – During the first quarter, protocols for extended V2I communication will be developed; and a procedure for mapping received BSM information for a given vehicle to its counterpart in the Surtrac predictive model will be determined.
[Months 4-6] – During the second quarter, procedures to update Surtrac predictive model and manage redundant messages will be completed, along with software to decode and interpret extended V2I messages to be received from CAVs.
[Months 7-9] – During the third quarter, a microscopic simulation of the Pittsburgh CV test bed will be developed, including mechanisms for simulating V2I communication and CAV nearby vehicle sensing, and an empirical analysis of traffic signal performance benefit (as outlined above) will be carried out. In parallel, work will commence on extracting and packaging information on nearby vehicles from Argo AI vehicle sensor systems.
[Months 10-12] – During the fourth quarter, a field implementation of CAV to Surtrac communication through respective OBU to RSU components will be completed, and further experimental validation will be carried out within the Pittsburgh CV test bed.
As indicated above, the proposed research includes an initial pilot test of capabilities in the field, using the current Pittsburgh CV test bed. This is viewed as the first step toward actual deployment. Assuming the pilot test is successful and indicates the viability of the approach, then both Argo AI and Rapid Flow Technologies would be natural candidates for hardening the technology and pushing it out for longer term deployment in Pittsburgh and other cities. One aspect of Argo AI’s plan for the future includes ride hailing services, and one would think that the ability to demonstrate a commitment to contributing to lessening congestion (in conjunction with Rapid Flow traffic control technology) could be viewed favorably by various municipalities.
Expected Accomplishments and Metrics
We expect the proposed ability to leverage CAV sensing capability to reduce uncertainty and improve the predictive modeling accuracy of adaptive traffic signal systems to translate directly to better traffic flow efficiency. The proposed experimental analysis will give insight into the extent of improvement that can be expected in terms of metrics that include average delay, average number of stops and average wait times. With the expected emergence of CAV operations over the next several years, we thus expect to significantly accelerate the longer term benefits that will be achievable with the future ubiquitous deployment of CV technology.
||Student - PhD
||Faculty - Researcher/Post-Doc
||Faculty - Tenured
Amount of UTC Funds Awarded
Total Project Budget (from all funding sources)
|Data Management Plan
||July 11, 2019, 5:22 a.m.
||March 27, 2020, 4:59 a.m.
||March 27, 2020, 4:59 a.m.
||Learning Model Parameters for Real-Time Traffic Signal Optimization
||Sept. 25, 2020, 6:36 a.m.
||Cooperative Schedule-Driven Intersection Control with Connected and Autonomous Vehicles”, Proceedings IEEE/RSJ International Conference on Intelligent Robots and Systems
||Sept. 25, 2020, 6:37 a.m.
||Multi-agent Sensor Fusion for Connected and Autonomous Vehicles to Enhance Navigation Safety
||Sept. 25, 2020, 6:56 a.m.
||Smart Infrastructure for Future Urban Mobility
||Sept. 25, 2020, 6:57 a.m.
||Multiagent sensor fusion for connected & autonomous vehicles to enhance navigation safety.
||Oct. 24, 2020, 7:26 p.m.
||Feb. 3, 2021, 6:57 p.m.
||Feb. 4, 2021, 5:04 a.m.
No match sources!
||Deployment Partner Deployment Partner
|Rapid Flow Technologies
||Deployment Partner Deployment Partner