Risk Aware Reservoir Control for Safer Urban Traffic Networks

Alexander Hammerl, Corresponding Author [email protected] Wen-long Jin [email protected] Ravi Seshadri [email protected] Thomas Kjær Rasmussen [email protected] Otto Anker Nielsen [email protected] · 2025

Read Risk Aware Reservoir Control for Safer Urban Traffic Networks by Alexander Hammerl, Corresponding Author [email protected] Wen-long Jin [email protected] Ravi Seshadri [email protected] Thomas Kjær Rasmussen [email protected] Otto Anker Nielsen [email protected] (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-06790.txt.

Introduction

Alexander Hammerl and four co-authors designed in 2025 a traffic controller that treats crash risk as a first-class goal for city networks. Their scheme squeezes a road network into two linked reservoirs whose flows follow a bathtub model. Crashes arrive as a self-exciting process whose rate grows with exposure, speed gaps and clustering. Each crash then drags down speeds and capacity in proportion to the live accident load.

A rolling-horizon controller balances delay against a risk penalty set by one dial. The math shows the best gates switch only when crashes occur, which makes updates event-triggered and cheap. Calibrated on Copenhagen map data for the morning peak, simulations cut delays by up to 30 percent and crashes by up to 35 percent. The authors present the trade-off between speed and safety as one transparent knob.

Catalog record

CREATORAlexander Hammerl, Corresponding Author [email protected] Wen-long Jin [email protected] Ravi Seshadri [email protected] Thomas Kjær Rasmussen [email protected] Otto Anker Nielsen [email protected]
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE62 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-06790

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Risk Aware Reservoir Control for Safer Urban Traffic Networks Alexander Hammerl, Corresponding Author [email protected] Wen-long Jin [email protected] Ravi Seshadri [email protected] Thomas Kjær Rasmussen [email protected] Otto Anker Nielsen [email protected] Abstract \AuthorHeaders Hammerl, Jin, Seshadri, Rasmussen and Nielsen 1 Abstract We present a risk-aware perimeter-style controller that couples safety and efficiency targets in large, heterogeneous urban traffic networks. The network is compressed into two interacting “reservoirs” whose dynamics follow the Generalized Bathtub Model, while accidents are described by a self-exciting (Hawkes) counting process whose intensity depends on vehicle exposure, speed dispersion between reservoirs and accident clustering. Accident occurrences feed back into operations through an analytically simple degradation factor that lowers speed and discharge capacity in proportion to the live accident load. A receding-horizon policy minimizes a mixed delay–safety objective that includes a variance penalty capturing risk aversion; the resulting open-loop problem is shown to possess a bang-bang optimum whose gates switch only at accident times. This structure enables an event-triggered MPC that only re-optimizes when new accidents occur, reducing on-line computation significantly. Parameters are calibrated using OpenStreetMap data for metropolitan Copenhagen to analyze traffic dynamics during morning peak commuter demand. Monte-Carlo simulations demonstrate delay savings of up to 30 % and accident reductions of up to 35% relative to an uncontrolled baseline, with a transparent trade-off governed by a single risk parameter. Keywords : Model Predictive Control, Generalized Bathtub Model, Traffic Safety, Incident Prevention, Self-Exciting Point Process 2 Introduction Road crashes remain a pressing public-health problem and a sizeable economic burden. The World Health Organization [ 1 ] estimates that they claim close to 1.3 million lives each year, while injuring between 20 and 50 million more, and societal costs are estimated in the high billions annually for major industrialized countries. Beyond these aggregates, accidents trigger secondary congestion, lost productivity and long-term welfare impacts that weigh heavily on urban economies. Urban areas are disproportionately affected. Dense road grids, frequent intersections and multimodal interactions magnify the secondary delays that follow an incident; even a minor crash may propagate queue spill-back across several signal cycles, disrupt public transport reliability and trigger follow-up collisions. Such systemic knock-on effects make it difficult to isolate the “pure” safety component from background congestion and call for an integrated modeling framework that treats accidents and traffic dynamics in a unified manner. Traffic flow theory provides the foundation for modeling vehicular dynamics and evaluating the effectiveness of accident response measures. Traditional approaches in traffic flow theory focus on spatiotemporal dynamics along individual road segments. This comprises both macroscopic frameworks that treat traffic as a continuous medium [ 2 , 3 , 4 , 5 , 6 ] and microscopic car-following models that capture individual vehicle behavior [ 7 , 8 , 9 , 10 ] . For network-level accident analysis, reservoir-based traffic flow models offer a complementary perspective by describing the dynamics of spatially aggregated traffic quantities across entire networks [ 11 , 12 , 13 , 14 , 15 ] . These models typically assume a time-invariant relationship between vehicle accumulation and average network flow, known as the Network Fundamental Diagram (NFD). Advanced reservoir frameworks also incorporate relationships between network accumulation and trip completion rates [ 16 , 17 , 18 , 19 ] , while the Generalized Bathtub Model [ 20 , 21 , 22 ] further extends this approach by accounting for the distribution of remaining trip lengths among individual vehicles. This spatial extension proves particularly relevant for understanding distributional effects that influence effective accident mitigation strategies in complex urban networks. On the other end of the spectrum, research on accident modeling and prevention spans a wide spectrum of temporal and spatial aggregation, and the relationship between traffic operations and safety has been a longstanding concern in transportation research. At the facility level, Safety Performance Functions (SPFs) have been widely used to predict crash frequencies based on long-term aggregates of daily traffic volumes and roadway features [ 23 , 24 , 25 , 26 ] . While such models typically report a positive correlation between traffic volumes and crash counts, this association may not persist under real-time or spatially disaggregated conditions. For example, positive correlations have also been found between average density indicators (such as the volume-to-capacity ratio) and crash frequencies [ 27 , 28 , 29 ] , while findings regarding the role of average speed remain mixed [ 30 , 31 , 32 , 33 , 34 , 35 ] . Another stream of safety literature investigates the influence of geometric design and intersection typologies. Roundabout studies such as [ 36 ] and [ 37 ] have demonstrated substantial safety gains, often independent of speed effects. [ 38 ] and [ 39 ] further developed accident prediction models highlighting the interplay between geometric complexity, vehicle familiarity, and traffic flow. At signalized intersections, dilemma-zone behavior [ 40 ] introduces unique risks, which are expected to be mitigated by the rule-based behavior of Connected and Autonomous Vehicles (CAVs). Microsimulation-based safety assessments have enabled richer representations of conflict interactions. [ 41 ] , [ 42 ] , and [ 43 ] used VISSIM and the Surrogate Safety Assessment Module (SSAM) to evaluate the trajectory-based safety performance of CAVs. These studies confirm that CAV operation reduces surrogate risk measures under various conditions, especially in platooning and mixed-fleet scenarios. [ 44 ] proposed a data-driven framework to estimate lane-change-related crash risk based on naturalistic driving data, while [ 45 ] demonstrated the power of high-fidelity crash simulations calibrated with real-world data for urban crash occurrence modeling. These are complemented by qualitative analyses such as [ 46 ] , which emphasize CAVs’ potential to prevent intersection-related conflicts. Machine learning-based approaches have also gained traction in real-time crash prediction. [ 47 ] compared shallow and deep learning architectures to forecast crash likelihood from traffic sensor data, illustrating the potential of data-driven models in operational safety monitoring. [ 48 ] proposed a multi-agent framework (FAIRLANE) for adaptive lane prioritization based on safety metrics in mixed autonomy traffic. Relatedly, [ 49 ] reviewed broadcasting strategies for vehicular networks, highlighting how timely dissemination of hazard information can help mitigate collision risk. At the network level, the Macroscopic Safety Diagram (MSD) framework introduced by [ 50 ] integrates operational state variables (e.g., mean network density) with crash propensity. These studies reveal that crash risk tends to peak in congested conditions—often at densities above the critical value that maximizes flow. Theoretical models based on the two-fluid theory have been proposed to support such observations [ 51 , 52 ] , although their applicability to urban network dynamics remains limited. Our work extends the existing literature by developing a network-level control framework that addresses accident risk in metropolitan areas holistically while remaining practical for real-world deployment. The framework requires only aggregate flow measurements and incident reports—data routinely available to traffic management agencies—and captures feedback from live accidents through an analytically simple capacity degradation term. This yields closed-form optimal policies with a bang-bang structure that switches only at accident events, enabling both conceptual transparency and computational efficiency suitable for real-time implementation. The rest of the paper is structured as follows. Section 2 presents the model description, introducing the Generalized Bathtub Model for traffic dynamics and the self-exciting Hawkes process

Read the complete text