Decentralized Routing on Spatial Networks with Stochastic Edge Weights

We investigate algorithms to find short paths in spatial networks with stochastic edge weights. Our formulation of the problem of finding short paths differs from traditional formulations because we specifically do not make two of the usual simplifying assumptions: (1) we allow edge weights to be stochastic rather than deterministic; and (2) we do not assume that global knowledge of a network is available. We develop a decentralized routing algorithm that provides en route guidance for travelers on a spatial network with stochastic edge weights without the need to rely on global knowledge about the network. To guide a traveler, our algorithm uses an estimation function that evaluates cumulative arrival probability distributions based on distances between pairs of nodes. The estimation function carries a notion of proximity between nodes and thereby enables routing without global knowledge. In testing our decentralized algorithm, we define a criterion that allows one to discriminate among arrival probability distributions, and we test our algorithm and this criterion using both synthetic and real networks.

💡 Research Summary

The paper tackles the classic shortest‑path problem under two realistic relaxations: edge weights are treated as stochastic variables rather than fixed numbers, and the routing entity does not possess complete, global knowledge of the network topology or its weight distributions. This setting reflects real‑world conditions such as traffic congestion, weather‑induced delays, or fluctuating communication latencies, where the exact cost of traversing a link can only be described by a probability distribution and where gathering full network information is often infeasible.

To address these challenges, the authors propose a decentralized routing algorithm that relies solely on locally available information and a distance‑based estimation function. The estimation function takes as input the Euclidean (or geodesic) distance between any two nodes and a set of pre‑learned statistical parameters describing the stochastic edge weights. It outputs a cumulative arrival‑time distribution for traveling between the two nodes, effectively quantifying the “proximity” of a neighbor to the destination in probabilistic terms.

The algorithm proceeds iteratively. At each step the current node evaluates all adjacent candidates. For each candidate, it combines the cumulative distribution already accrued along the path so far with the distribution predicted by the estimation function for the remaining segment to the destination. The candidate that maximizes the probability of arriving at the target within a predefined time horizon (or, equivalently, that yields the highest cumulative arrival probability, CAP) is selected as the next hop. After the hop is taken, the observed travel time on the traversed edge is used to update the estimation function, allowing the method to adapt to real‑time fluctuations.

The authors test the approach on two classes of networks. The first is a synthetic grid where edge travel times follow independent Gaussian distributions with known means and variances. The second consists of a real‑world urban road network, for which empirical speed and variance data are extracted from traffic sensors. In both cases, performance is measured by the CAP metric, i.e., the probability of reaching the destination before a deadline. Compared with classic deterministic algorithms such as Dijkstra’s or A* (applied to the expected edge costs), the decentralized stochastic method consistently achieves higher CAP values, especially when edge‑weight variability is large or when the network size grows.

A further contribution is the definition of a CAP‑based selection criterion. Instead of minimizing expected travel time alone, the criterion evaluates full arrival‑time distributions and selects the path that offers the best trade‑off between expected speed and risk of delay, often using a confidence‑level threshold (e.g., 90 % reliability). This risk‑aware perspective is more appropriate for applications where late arrival carries a high penalty.

The paper also discusses limitations. The estimation function requires an initial training phase with sufficient historical data to capture the statistical properties of edge weights. Sudden, extreme events (e.g., accidents) may temporarily degrade prediction accuracy until enough new observations are incorporated. The authors suggest future work on online learning, multi‑objective extensions (e.g., energy consumption), and integration with multimodal transport or communication networks.

In summary, the study presents a novel framework for routing on spatial networks where edge costs are uncertain and global knowledge is unavailable. By leveraging a distance‑driven probabilistic estimator and a CAP‑centric decision rule, the algorithm provides real‑time, risk‑aware guidance that outperforms deterministic baselines under realistic stochastic conditions. This contribution has immediate relevance to smart‑city traffic management, autonomous vehicle navigation, and dynamic routing in volatile communication infrastructures.