Design analysis of a decentralized equilibrium-routing strategy for intelligent vehicles
Transportation Research Part C: Emerging Technologies
You share, we take care!' -Taverne project https://www.openaccess.nl/en/you-share-we-take-care Otherwise as indicated in A B S T R A C T Intelligent vehicle technologies are opening new possibilities for decentralized vehicle routing systems, suitable for regulating large traffic networks, and at the same time, capable of providing customized advice to individual vehicles. In this study, we perform a rigorous simulation-based analysis of an in-vehicle routing strategy that aims to achieve a
... -equilibrium distribution in traffic. Novel features of the approach include: a mechanism based on forward propagation of individual vehicle decisions to anticipate future traffic dynamics; time-dependent prediction of route travel times with neural network-based link predictors; and a stochastic routing policy for invehicle decision-making based on predicted travel times. However, for an effective application of the approach, design choices need to be made regarding the accuracy of the link predictors, and some control settings. These choices may depend on the network size and structure. We investigate the impact of two important design aspects: sequentially using link-level predictors for route travel time estimation, and the control parameter values, on the equilibrium performance at the networklevel. The results suggest functional scalability of the approach, in terms of the prediction model accuracy and routing performance. Overall, the work contributes to a qualitative and quantitative understanding of emergent performance from the given routing approach. Introduction Dynamic route guidance in freeway or urban traffic networks with time-varying demand and stochastic travel times is a classic transportation problem (Papageorgiou, 1990) . At the same time, recent advances in automated and connected vehicle capabilities (Shladover, 2017), boosted by other emerging technologies, like cloud computing, artificial intelligence, big data, and internet of things, are fundamentally transforming the potential and design of traffic control systems Papageorgiou et al., 2015) . In-vehicle route guidance systems, be they satellite navigation devices or GPS enabled smartphone applications, such as Google Maps, Waze, and Apple Maps, are becoming ubiquitous and connected. These trends are making decentralized routing approaches increasingly more practicable. The benefits of a decentralized control structure range from lower computation and communication loads, higher fault tolerance, to robustness against measurement errors, delays and failures. Moreover, decentralized decision-making makes it possible to provide personalized advice to individual vehicles. We focus on decentralized route guidance systems that can achieve a user-equilibrium (UE) condition, famously known as the Wardrop's first principle of route choice T (Wardrop, 1952) . UE routing entails reallocating vehicle flows in a way that individual travel time costs are minimized, and cannot be further decreased by unilateral action. Not just is a UE optimal for individual users, it also results in a fair distribution, wherein all vehicles over all used routes between an origin-destination (OD) experience equal travel times. In this work, we study the design aspects of a cooperative route guidance system proposed by Claes (2015) for UE routing of individual vehicles. The investigated strategy uses a decentralized predictive algorithm, wherein the prediction model is decentralized at the level of individual links, and the routing decisions are decentralized at the level of individual vehicles. Central to its design is a coordination mechanism that does not use direct vehicle-to-vehicle communication. Instead, intelligent vehicles (IVs) cooperate by delegating virtual agents to share their planned (future) routes with roadside agents, which are computation and communication devices deployed across links in the physical network. These roadside agents constitute the environment that records information of individual route plans, and aggregate it to estimate time-dependent travel times. The vehicles can thereafter use the predictions to update their future routing plans. In this way, by integrating the travel time prediction and routing models, not just the current traffic condition but also the control decisions can be included in the predictions. An effective implementation of the approach depends on the understanding of some important design aspects. First, the timedependent route travel time predictions are made in a decentralized model using multiple Artificial Neural Network (ANN)-based link predictors. Individual link predictors estimate link travel times using local in-vehicle information. These link predictions are then aggregated to estimate route travel times. The issue however is that the errors in link estimates may negatively impact the prediction accuracy of subsequent link predictors. Thus, the propagation of error over multiple links determines the reliability of the route travel time predictions. How the link and route travel time accuracy relate, is moreover valuable for identifying a design criterion for the link predictor accuracy. Next, individual vehicles instantaneously react to these predictions using a stochastic routing decisioncriterion. This is unlike model-predictive type control where the control signal is explicitly optimized over a future horizon. The routing behavior thus achieved depends on the choice of two control parameters -one of which influences the responsiveness of the routing decision to the travel time difference between the chosen and the fastest alternatives, and the other specifies the frequency of decision update. Tuning these control parameters requires identifying mechanisms and quantities that reflect the impact of the parameter values on the system performance. We study the discussed design aspects using systematic simulations, making the following research contributions: N. Mahajan, et al.