U.S. Department of Transportation
Federal Highway Administration
1200 New Jersey Avenue, SE
Washington, DC 20590
202-366-4000
Federal Highway Administration Research and Technology
Coordinating, Developing, and Delivering Highway Transportation Innovations
| SUMMARY REPORT |
| This summary report is an archived publication and may contain dated technical, contact, and link information |
|
 |
| Publication Number: FHWA-HRT-15-085 Date: December 2015 |
Publication Number: FHWA-HRT-15-085 Date: December 2015 |
David Gibson at FHWA began the second day of the workshop with a brief summary of the previous day's presentations and discussions. He noted that control-system layouts of the past are now obsolete because the sensor landscape has changed over time to include technology such as Bluetooth™ sensors, vehicle reidentification, and dedicated short-range communications. Gibson mentioned that data fusion is an important part of improving algorithms and, for control systems, it is important to understand what to optimize. He suggested that, for both sensors and control algorithms, it is important to have a simulation and optimization framework to support research.
Gibson continued to explain that researching and developing new traffic-signal control algorithms requires a fairly complex simulation infrastructure. The demands of assembling the infrastructure often lead researchers to spend more time on the infrastructure than on the core question of control algorithm development. Challenging issues with intellectual property rights also remain, suggesting the need for open source approaches. The second day of the workshop aimed to accelerate the development of tools for future signal control research by confirming the need for and potential benefits of these tools and considerations for their development and dissemination.
Presentations and facilitated discussion from the second day of the workshop are summarized in the following pages.
Dr. Larry Head
Associate Professor of Systems and Industrial Engineering
and Engineering Management, University of Arizona
Dr. Head presented several challenges to workshop participants that are associated with traffic control systems. These challenges are outlined as follows:
Challenge 1: Configuration Management
Dr. Head highlighted that configuring traffic-signal control systems properly is critical to getting them to perform properly; however, he also noted that there is not a reliable way to know which intersections are configured poorly. The current method requires a test vehicle to actuate the detector in the field. Dr. Head suggested that this is not a cost-effective method and that there must be a better way to move forward. He noted that personnel need to be trained to configure the systems properly and certification might be necessary. Dr. Head also mentioned that security is an issue and will be especially important as connected vehicles roll out.
Challenge 2: Detection
Dr. Head noted that practitioners need to not only know how to assess and use detection data but also how to make detectors more reliable and available. He suggested that researchers should think differently about detection methods for intersections and freeways. Dr. Head mentioned that different technologies could be explored, such as light detection and ranging and ultrasound. He highlighted that video technologies at intersections could also give good situational awareness and clear images of the environment. Dr. Head emphasized that different technologies are useful for different objectives.
Challenge 3: Performance Measurement
Dr. Head suggested that there may be more useful performance measures beyond the traditional methods, such as delays, stops, and travel time. He mentioned that a useful metric could be to measure the number of controllers or detectors that are working properly, or to measure availability, reliability, and quality. Dr. Head noted that measuring delay and travel time does not indicate how well the system is functioning. He also noted that many different metrics are required to operate the system, such as turning-movement counts, phase-demand volume, and physical geometry.
Challenge 4: Control Algorithms
Dr. Head noted that practitioners need to maximize the usefulness of their controllers. He described how traffic controllers have many features that are highly specialized, but operators do not always know about all of the different features. Dr. Head provided the example of how fixed-time control works well in a central business district but it is not necessarily good for other situations. In addition, Dr. Head mentioned that actuated and coordinated actuated signals work well to minimize phase failures but are not effective in congested conditions. These signals also depend on the detection system reliability. Dr. Head highlighted that there are a lot of parameters to consider and that they must be configured properly to optimize performance.
Challenge 5: Multimodal Control
Dr. Head highlighted that there are many different users of the transportation system, each with their own set of requirements, which result in competing interests. He noted that most traffic control systems treat modes other than passenger vehicles as local exceptions. As an example, Dr. Head reminded workshop participants that pedestrians have to push a button to actuate the signal. He noted that coordinating all of the modes in practice on a corridor scale is rare and difficult.
Dr. Head proposed several research topics, including longer term and higher risk research topics, as follows:
In summary, Dr. Head reminded workshop participants that adaptive controllers have the capability to learn how the intersection works instead of relying on pre-programming. He highlighted that there are too many parameters to adjust and features that are specific to special situations or particular modes. Dr. Head suggested that this makes configuration so complicated that most people do not understand how to do it well. He noted that controller configuration could be made simpler.
Dr. Steve Remias
Transportation Research Engineer, Department
of Civil Engineering, Purdue University
During his presentation, Dr. Remias described the evolution of transportation operation from fixed time, to actuated, to adaptive control.
Dr. Remias discussed several themes during his presentation. He noted that there is a misconception that adaptive is a "set it and forget it" controller. He suggested that the situation is more complicated than that. Dr. Remias highlighted that performance measures are complex and need to be actionable and that there is no single measure that can tell us everything. He mentioned that a set of performance measures will provide a comprehensive picture and that these performance measures must go back into the system and be reinforced when performance is good.
Dr. Remias also discussed signal timing and noted that it is a six-step process. He focused on the final step in this process, which is the assessment step. This refers to the types and quality of data that are used. Dr. Remias noted that the results of the assessment will provide feedback for the system. He also noted that high-resolution signal-control data capture can collect data any time the signal state changes or if a vehicle is detected. Dr. Remias mentioned that this is very useful. He noted that all of the controller manufacturers at this time implement this in different ways but that these differences are still being worked out. He also mentioned that it is now possible to embed equipment that collects high-resolution signal-control data.
Dr. Remias highlighted several relevant projects at Purdue University, as follows:
At the conclusion of his presentation, Dr. Remias identified some near-term research opportunities, as follows:
Figure 5. A coordination diagram developed by Purdue University.
Note: black dots represent vehicle arrivals; the different phases of the cycle are represented by red and green.
Dr. Pitu Mirchandani
Professor of Computing, Informatics, and
Decision Systems Engineering, Arizona
State University
Dr. Steve Shelby
Senior Research Engineer,
Econolite Group, Inc.
MODERATOR
Raj Ghaman
Texas Transportation Institute
The next two presentations provided perspectives from academic and industry representatives involved in adaptive traffic control.
Dr. Mirchandani explained that when there is little fluctuation between supply and demand in transportation, the system moves toward equilibrium. He noted that this is rarely the case and suggested that fluctuations can be dealt with either proactively or reactively. Dr. Mirchandani suggested that sensors, communication infrastructure, central processing units, and engineers are necessary to be proactive. Dr. Mirchandani also mentioned one of his projects that streams real-time field data to a bank of servers in a cloud and makes predictions and adjustments by using a cyber-physical system, known as managing interactive demands and supplies (MIDAS). He noted that MIDAS uses big data, real-time algorithms, and cloud computing to proactively manage traffic, as shown in figure 6.
Figure 6. An overview of managing interactive demands and supplies.
Dr. Mirchandani explained that, in proactive control, sensors are used to measure the outputs of the system, and then models perform the estimation. He mentioned that decisions are made and controls are based on these decisions. He highlighted several features of proactive traffic control, including that it:
Dr. Mirchandani noted that MIDAS focuses on lanes instead of links. He explained that fluid flow, in relation to traffic-flow theory, does not consider that freeways have lanes that may have different types of vehicles (e.g., automated or connected). He suggested that predictions must be made about what is happening in lanes and that this can produce lane-based real-time speed maps, which are better than aggregate predictions and allow for managed lane predictions.
Dr. Mirchandani described a control system that he developed called the real-time hierarchical optimized distributed effective system (RHODES). He highlighted that the system proactively predicts demand at traffic signals. It is based on a real-time feedback control paradigm, will work with multiple modes, and is applicable to intersections, arterials, and networks. Dr. Mirchandani explained that RHODES uses discharge rates, travel times, and turn ratios to predict arrivals and queues, which then feed a control algorithm. He described how this algorithm, known as the categorized arrivals-based phase re-optimization at intersections control algorithm, is a real-time algorithm that determines the duration of the different phases of the signal, allows various objectives (e.g., optimization of the delay) to be programmed for different vehicle classes, considers categories of predicted arrivals and their objectives, and considers a given rolling decision time horizon with time increments in seconds. Dr. Mirchandani noted that RHODES enables a user to compute total delay and stops over an entire corridor with many intersections.
Dr. Mirchandani also introduced workshop participants to the next generation of RHODES, known as self-tuning RHODES. He explained that the performance of RHODES is directly related to the accuracy of its queue estimates, which are dependent on parameters that are not fixed (e.g., turn proportions, queue-discharge rates, and link-travel times).
At the conclusion of his presentation, Dr. Mirchandani noted several general research observations, as follows:
He also made the following concluding remarks:
Dr. Shelby explained that traffic management systems in the future will be able to obtain better information about vehicles in the network, which will enable more capable adaptive control. He noted that connected vehicles will arrive soon, broadcasting real-time location information, and suggested that this will transform how optimization algorithms work.
Dr. Shelby highlighted that it was approximately 20-years ago that FHWA unveiled the real-time traffic adaptive control system program. He reminded workshop participants that it is important to understand key findings from adaptive control research in the past 20 years, in both academia and industry. He highlighted that this informed perspective suggests more practical expectations for the future. Dr. Shelby noted that the work of adaptive control proved to be a challenging multidisciplinary endeavor. He mentioned that, of the five groups funded by FHWA under the real-time traffic adaptive control system program, three groups produced functional software. Of these three prototypes, one demonstrated performance benefits in an independent traffic simulation evaluation, and none produced significant performance benefits versus traditional time-of-day control in field operational tests. Dr. Shelby noted that when traffic was not heavy, there were no pedestrians, and the system was unconstrained in terms of phase sequencing, the real-time adaptive signal control system performed very well. However, he highlighted that the system did not perform well when traffic was heavy, if there were regular pedestrians actuating "WALK" signal timing, if the adaptive control system was constrained in terms of the phase sequence, or if it was not allowed to skip phases based on demand from pedestrians or vehicles.
Dr. Shelby suggested a need to consider longer time horizons for optimizing traffic flow. He mentioned that planning a full cycle ahead (in this case, over 100 seconds) is a challenge for acyclic adaptive control systems, and computational effort grows very fast as a function of the optimization horizon. Unlike cyclic systems, which distribute time cyclically to each phase of the intersection, acyclic systems decide every time whether to switch or not to switch to the next phase. Dr. Shelby highlighted that perfect information does not overcome the fundamental challenges of a problem and, while it helps, it does not yield a paradigm leap over traditional control. He noted that, even with perfect information, the computational complexity of the optimization still requires timesaving optimization shortcuts that can degrade performance. Dr. Shelby also mentioned that predicting arrivals on the order of a large cycle length is still a problem because those vehicles still have to drive through several other acyclic adaptive signals before they will arrive.
He noted that agencies are very cost sensitive and will not adopt new technology unless it is cost-effective and adds value. He also noted that agencies did not embrace adaptive control systems because they were expensive, complex, and did not consistently improve traffic flow relative to traditional nonadaptive signal timing. He suggested that, because agencies are so sensitive to cost, value considerations may influence perspectives on what types of future research might be more practical and valuable. He highlighted that even with a mandate for all newly manufactured vehicles to broadcast location, there would still be a requirement for infrastructure-based sensors to see and serve "unconnected" vehicles, bicycles, and pedestrians.
Dr. Shelby highlighted that the next generation of sensors will provide object-tracking capability instead of just zone-presence capability. He mentioned that, whether it is through light detection and ranging, radar, or another technology, the next generation is likely to overcome some of the detection problems attributed to night time, glare, shadows, fog, or other issues that hamper video detection today. He suggested that adaptive control could benefit from more accurate sensors but, as referenced previously, even perfect information is not going to substantially change performance at typical intersections during typical conditions.
Edward Fok
Transportation Technology Specialist,
FHWA Resource Center
Dr. Douglas Gettman
Systems Manager,
Kimley-Horn and Associates, Inc.
MODERATOR
Deborah Curtis
Federal Highway Administration
The researchers of the next two presentations discussed tool considerations for advanced traffic control systems and tools for perfor mance assessment and agency application development issues.
Edward Fok suggested several requirements for successful traffic control. He mentioned that it is important to consider end users and their goals, limitations, and concerns. Fok noted that safety is a goal that is the combined responsibility of the vendor and the agency. Operators need to make sure that their systems are used correctly, and this places constraints on the system. He noted that safety needs to be guaranteed by vendor, manufacturer, and software supplier. Fok highlighted that mobility is another goal that requires a lot of data to meet. He suggested that there needs to be a case made for keeping and using the data so that agencies prioritize doing so. Fok also highlighted that accountability is important to agencies, so that they know that a tool is doing what it is supposed to be doing. He noted that the privacy issues related to large datasets also need to be resolved.
Fok also suggested some capabilities and limitations for workshop participants to consider, as follows:
Fok also highlighted that the return on investment needs to be understood for safety, mobility, and staff investment and metrics need to be established to show this return.
In the next presentation, Dr. Gettman discussed some of the barriers to adaptive control, including cost and uncertainty of benefits. He mentioned that previous work addressed some of the cost-barrier issues. He also highlighted a current effort with FHWA, as part of the Every Day Counts initiative, to address some of the issues with respect to uncertain benefits. He noted that the goal is to get proven technologies used more often in industry, and adaptive signal control was one of the selected programs. Dr. Gettman attributed uncertainty of benefits to the following factors:
Dr. Gettman noted that adaptive control can sometimes produce false good results. He mentioned that this could happen because the inputs are bad, the results are cherry picked, or if the traffic is changing. He also noted that at other times the adaptive control performance may influence results.
Dr. Gettman highlighted that industry is developing tools for deploying adaptive systems. These tools include Bluetooth™ sensors, smart phones, traffic counters, and controllers to collect data that are processed to generate MOEs. Dr. Gettman mentioned several available MOEs, includin percentage of arrivals on green, platoon ratio, green-occupancy ratio, reliability, throughput, route-travel time, route-travel delay, travel-time reliability, and stops per mile. He also mentioned some MOEs for future use, including emissions, regional aggregates, charts or graphs, and dashboards.
At the conclusion of his presentation, Dr. Gettman outlined some additional thoughts, as follows:
At the conclusion of the workshop, David Kuehn, EAR Program Manager, facilitated a group discussion about what was learned at the workshop. The key discussion points and final thoughts of workshop participants are summarized as follows:
