Periodic Reporting for period 1 - MAGnUMplus (Optimal urban traffic routing by broadcasting avoidance maps)
Periodo di rendicontazione: 2020-12-01 al 2022-05-31
The PoC developed a prototype for an optimal route guidance system that improves traffic conditions in urban areas. Urban traffic management at a large scale is challenging but may lead to significant travel time savings by better-distributing drivers among the network. Existing navigation apps or routing systems provide the shortest path in time to users resulting in the network user equilibrium. However, traffic engineers know that total travel times may be reduced by 10 to 30% if user routes comply with the system optimum. There is no actual traffic management system that can achieve such a goal because of computational (determining the optimal route for all current users in NP-hard), privacy (optimality requires that all users share their destination with a centralized controller), and compliance issues (users may not follow routing instructions).
The optimal route guidance system we have designed within the MAGnUM project can solve the two first issues. A centralized controller produces real-time avoidance maps, i.e. the definition of how many users should avoid each subsection of the road network to alleviate congestion in this area. Such maps are derived by monitoring overall traffic conditions. Each user transforms this information into individual route guidance through its navigation system. This privacy is guaranteed by design as the users do not share information with the controller but only benefit from avoidance maps.
The PoC demonstrated that all required components for the system to operate in real-time at large-scale work. The critical one is the local speed estimation in the different monitored regions because not enough observations may be available for some areas at a given time. We have developed a new method based on neighboring regions and historical data to circumvent this issue. However, it appears that data providers are doing data filtering before publishing the data in real-time. Such a step may introduce bias in the regional speed estimation. To further develop the solution, it is essential to look for partnerships with traffic data providers to secure accurate speed estimates. Concerning the other components/implementation steps, network partitioning needs more research to automatize the process better. Still, we proposed during this PoC a new method based on the network properties (node betweenness centrality), which is a significant step toward full automatization. The PoC showed that calculating optimal route guidance for many users in real-time is straightforward.
The main focus of the PoC was to investigate users’ reactions to the guidance and determine the natural compliance rate. We concluded that users are willing to change their route to improve the overall traffic conditions if their effort remains small, i.e. no more than 40% of extra traveled distance compared to their regular trip (without the avoidance maps). 30% of the users do not comply, while 20% strictly follow the recommended routes. In between, we observe a lot of partial compliance. On average, when the requested effort is below 40%, the users will extend their distance traveled by 70% of the required effort. More importantly, if we consider the distance traveled in the avoided regions when the system is running, with the same distance and the system off, we found a ratio of 70%. That means that the system based only on the natural user acceptance can reduce the distance traveled in the congested (avoided) regions by 30%. It is a significant number considering the individual effort users must make, but it may be insufficient for the system to work at total capacity. Incentives or penalties enforced by the local authorities may be an excellent addition to reaching optimal performance in the field.
We are now going to further investigate the system performance by simulation, considering what we learned about the user behaviors and work on the methods to ease the implementation in a new city. We will look for industrial partners willing to develop the prototype further and run new experiments.
The optimal route guidance system we have designed within the MAGnUM project can solve the two first issues. A centralized controller produces real-time avoidance maps, i.e. the definition of how many users should avoid each subsection of the road network to alleviate congestion in this area. Such maps are derived by monitoring overall traffic conditions. Each user transforms this information into individual route guidance through its navigation system. This privacy is guaranteed by design as the users do not share information with the controller but only benefit from avoidance maps.
The PoC demonstrated that all required components for the system to operate in real-time at large-scale work. The critical one is the local speed estimation in the different monitored regions because not enough observations may be available for some areas at a given time. We have developed a new method based on neighboring regions and historical data to circumvent this issue. However, it appears that data providers are doing data filtering before publishing the data in real-time. Such a step may introduce bias in the regional speed estimation. To further develop the solution, it is essential to look for partnerships with traffic data providers to secure accurate speed estimates. Concerning the other components/implementation steps, network partitioning needs more research to automatize the process better. Still, we proposed during this PoC a new method based on the network properties (node betweenness centrality), which is a significant step toward full automatization. The PoC showed that calculating optimal route guidance for many users in real-time is straightforward.
The main focus of the PoC was to investigate users’ reactions to the guidance and determine the natural compliance rate. We concluded that users are willing to change their route to improve the overall traffic conditions if their effort remains small, i.e. no more than 40% of extra traveled distance compared to their regular trip (without the avoidance maps). 30% of the users do not comply, while 20% strictly follow the recommended routes. In between, we observe a lot of partial compliance. On average, when the requested effort is below 40%, the users will extend their distance traveled by 70% of the required effort. More importantly, if we consider the distance traveled in the avoided regions when the system is running, with the same distance and the system off, we found a ratio of 70%. That means that the system based only on the natural user acceptance can reduce the distance traveled in the congested (avoided) regions by 30%. It is a significant number considering the individual effort users must make, but it may be insufficient for the system to work at total capacity. Incentives or penalties enforced by the local authorities may be an excellent addition to reaching optimal performance in the field.
We are now going to further investigate the system performance by simulation, considering what we learned about the user behaviors and work on the methods to ease the implementation in a new city. We will look for industrial partners willing to develop the prototype further and run new experiments.