As traffic engineers, our ability to improve traffic signal timing hinges upon our ability to accurately measure signal performance metrics. It’s critical that we understand factors like travel time, throughput, control delay, proportion of vehicles arriving on green, and so on. Thanks to considerable improvements in traffic detection technology, accessing accurate performance metrics is becoming more and more achievable. However, these metrics typically rely on costly hardware and maintenance. Axilion is using a unique methodology to produce signal performance metrics which uses its proprietary digital twin environment and requires no additional hardware.

What are Automated Traffic Signal Performance Measures (ATSPM)?

Automated Traffic Signal Performance Measures (ATSPM) are tools, like software platforms, that use recorded events from the controller¹ to automatically generate metrics and analytics. These metrics, often referred to as measures of effectiveness (MOEs), include factors like turning movement counts, wait times, and frequency of split failures. They provide valuable insight into the level of performance of a traffic signal.

According to the USDOT, approximately 26 transportation agencies at both state and local levels are currently involved in implementing ATSPMs. A great example of this is the Utah DOT’s open source software which allows the public to select a desired signal and performance measure to create charts and visualizations like the example below (see Figure 1).

Drawbacks in current methods of producing MOEs

For ATSPMs to produce quality MOEs, they require robust, accurate traffic data. This data is captured by detectors on the roadways like loop detectors, radar, and cameras. However, capturing the level of accuracy needed for ATSPMs can be a challenge due to the need for wide detector coverage and specific detector schemes.

Different use cases for traffic data require different capture methods. For instance, a traffic management agency needing to capture vehicle approaches at a traffic signal would place their underground detector ahead of the stop bar. This placement allows them to perform operational tasks, such as assigning right-of-way time intervals. However, for accurate data required for certain metrics, such as volumes, yellow actuations, and red light violations, the detector needs to be placed behind the stop bar.

This difference in required detection schemes presents a challenge for traffic management agencies, as detectors are a costly investment, from installation to ongoing server storage and maintenance requirements. Balancing the need for operational data and the precision required for ATSPMs often necessitates significant additional resources.

How Axilion produces MOEs with a digital twin

Unlike the traditional ATSPM tools, Axilion’s software, called X Way, leverages a digital twin to produce MOEs for evaluating signal performance. Its digital twin is a virtual environment that emulates a real-life roadway, mirroring traffic signal controller configurations and geographic layout (see Figure 2). Once injected with traffic data (and pedestrian data if available), the digital twin simulates vehicle movements inside the corridor, reflecting signal timing and other factors present in the actual environment. The simulated movement of vehicles throughout the virtual corridor allows X Way to produce a wide variety of MOEs to better understand traffic signal timing performance. These MOEs are captured using different methodologies described below.

Injection of traffic data

When traffic detection data is limited, which is often the case, Axilion’s methodology involves supplementing and meshing it with connected vehicle data. We source this data from partners like StreetLight, which provides historic turning movement counts at 15-minute intervals. This meshed data is later used to produce origin-destination (O-D) matrices which help simulate corridor traffic and commuter demand. The accuracy of our digital twin’s metrics are reliant on the data we can feed it. The more qualified and detailed data sources we have, the better the accuracy. 

1. Traditional charts using virtual detection: 

Purdue Coordination Diagrams (PCDs): By showing vehicle actuations in a spatial manner, PCDs provide a good way to analyze traffic progression and understand arrivals on green (AoG)². PCDs can also provide insight into possible opportunities for making improvements or corrections to offsets, cycle lengths, splits, etc.

X Way is able to produce PCDs (see figure 3) by simulating a setback detector within its digital twin. When a simulated vehicle passes over this setback detector, the detector captures the timestamp. This timestamp allows X Way to predict when the vehicle will reach the stop bar. Each predicted stop bar arrival is marked on the PCD with a black dot, where the X-Axis represents the timestamp and the Y-Axis indicates the cycle time. Additionally, the diagram includes the phases of the traffic signal—red, yellow, and green—to provide a comprehensive view of traffic signal timing.

Purdue Split Failure Diagram (PSFD): PSFDs present the green occupancy ratio and the red occupancy ratio during the first five seconds of red. When both are higher than 80%, this indicates a split failure³. Figure 4 shows an example of a PSFD in the X Way platform. The platform uses virtual detectors at the stopbar for all phases and movements to calculate occupancy for this chart creation.

In the PSFD, a split failure is indicated by the yellow vertical line. The diagram also uses specific shapes to represent the reasons for the end of the green interval: a triangle indicates a gap out⁴ and a circle signifies a force off⁵. An engineer can use this information to understand the cause of the split failure (i.e. a short split), and determine how to best address it (i.e. extend the passage time).  

Time-Space Diagrams: Time-Space Diagrams allow you to visualize corridor progression, showing the theoretical trajectory of a vehicle entering the corridor at a specific time and predicting its interactions with traffic signals. This diagram integrates inputs such as individual intersection locations, cycle length, splits, offsets, left turn phasing, and speed limit. It can be used to analyze a coordination strategy and modify timing plans.

Axilion can produce these diagrams when a user selects a 30-minute segment of time for a deeper analysis, called “microanalytics” in X Way. As Figure 5 shows, the larger green bandwidths represent greater opportunities for uninterrupted vehicle progression between adjacent intersections. Yellow bandwidths signify potential for a continuous green wave across the corridor, highlighting possible synchronization improvements.

2. Advanced analytics using probe data trajectory: 

While X Way uses virtual presence detectors to produce traditional charts, these are not essential for generating MOEs in the platform. X Way’s digital twin captures a vehicle’s trajectory in a manner similar to probe or floating car data, where vehicles are geo-localized and their movement is recorded. By using the digital twin, the platform can provide MOEs on travel times, speeds, arrivals on green, split failures, throughput, and much more. 

1. Arrivals on green: Within the platform, users can access the “Ng” value (see Figure 6), which represents the total number of arrivals on green across the coordinated phases. This value is available for 30-minute segments of time within X Way’s Micro-analytics page. The “Ng” value is derived from the digital twin by tracking the positions of simulated vehicles as they cross the stop bar.

2. Split failures: While the traditional method for identifying split failures uses the PSFD methodology, X way can inform users of split failures based on the number of cycles that each vehicle has spent waiting at the traffic light. This is possible due to the system’s ability to understand the signal intervals and the location of the simulated drivers in the digital twin. If the simulated vehicle has to wait for two or more cycles before managing to cross the intersection, this indicates a low level of service and is considered as a split failure.

Figure 7 shows the user interface presenting the percentage of drivers that are stopping at 0, 1, 2, and 2+ cycles.

3. Platoon ratio: X Way also produces Platoon Ratio charts, such as the example in Figure 8. This ratio is calculated based on the AoG values relative to green time. The values on the left side of the chart indicate the density of the platoon⁶: higher numbers represent high progression quality, while lower numbers (0.5 or less) indicate very poor progression quality. The greater the portion of vehicles that arrive on green using a smaller portion of the green time, the higher the platoon ratio and the better the quality of progression.

4. Control Delay: Control delay refers to the additional time vehicles spend at an intersection due to traffic signals. X Way provides control delay metrics for each vehicle group within the system (see Figure 10). This is calculated by comparing the predicted travel time—assuming the vehicle moves at the speed limit without any interruptions—with the average time recorded in the simulation as the vehicle moves through the corridor. The difference between these two times indicates the control delay.

Using a digital twin presents an innovative way for achieving MOEs for traffic signal performance without the need to install special detection mechanisms on the roadway. While traditional detectors are limited based on their fixed location, X Way’s digital twin presents a way to use probe trajectory data for a more detailed view of the vehicle’s movements. 

The examples I’ve presented are just a few ways that Axilion achieves this with its X Way platform. Users should look at this as a sandbox to be able to assess the MOEs resulting from different traffic signal timing plans without needing to make interventions in the physical controller until reviewing it in the digital twin environment first. 

Our future blogs will explore additional use cases for X Way’s digital twin, including AI-based optimization using reinforcement learning mechanisms. I will also delve into our unsupervised traffic clustering mechanisms for traffic pattern analysis.