⚡ electrical engineering

RoboSense: Leveraging Robotaxi Fleets as Drive-by Sensors for Urban Traffic Monitoring

This paper proposes a novel dynamic robotaxi routing framework that leverages fleet vehicles as cooperative drive-by sensors to simultaneously optimize traffic monitoring coverage and travel efficiency, demonstrating that integrating spatiotemporal coverage objectives can create a win-win scenario for both urban traffic management and robotaxi mobility.

Original authors: Yilin Wang, Yiheng Feng

Published 2026-09-09

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Yilin Wang, Yiheng Feng

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Traffic monitoring is the nervous system of a city, a constant effort to understand how vehicles move, where they get stuck, and how fast they are traveling. For decades, this job has fallen to fixed sensors buried in the road or cameras mounted on poles. These tools work well at specific spots, but they leave vast gaps in the picture, unable to see what is happening between them. More recently, cities have begun to use regular cars equipped with GPS as moving probes, but these vehicles are driven by individual needs, meaning they often crowd the same popular routes while leaving others in the dark. A new generation of self-driving taxis, known as robotaxis, offers a different possibility. Unlike private cars, these vehicles operate as a coordinated fleet under the control of a central manager. They are already equipped with powerful sensors to see the road, and they can share that data in real time. The question researchers asked was whether this fleet could do more than just carry passengers; could they be steered to become a mobile network of eyes, filling in the blind spots of traditional monitoring without slowing down the service they provide?

A team of researchers at Purdue University set out to answer this by creating a digital twin of a city to test a new way of routing these robotaxis. They built a simulation of a 5-by-5 grid network, a simplified urban layout with twenty-five intersections and forty road links, and filled it with background traffic to mimic a real city. Into this mix, they introduced a fleet of robotaxis at different levels of presence, ranging from two percent to ten percent of the total vehicles on the road. The core of their work was a computer program that decided where each robotaxi should go. Instead of simply sending every vehicle down the fastest path to its destination, the program was given a second goal: to maximize the amount of road space observed by the fleet. The researchers treated the road as a series of small, manageable blocks, and the program calculated how to move the fleet so that these blocks were covered as evenly as possible over time, ensuring that no part of the network was left unwatched for too long.

The results of these simulations revealed a surprising dynamic between the goal of moving passengers quickly and the goal of watching the road. Initially, the researchers expected that asking the fleet to monitor the city would force the vehicles to take longer, less efficient routes, slowing them down and frustrating passengers. However, the data showed something more nuanced. When the researchers balanced the two goals carefully, the fleet actually moved faster on average. This happened because the act of monitoring the road more thoroughly provided the system with better, more accurate information about traffic conditions. With a clearer picture of where traffic was building up, the routing computer could predict travel times more precisely and guide the vehicles along paths that avoided congestion. In this way, the effort to watch the city improved the ability to drive through it, creating a situation where both monitoring quality and travel speed increased together.

This improvement, however, had limits. The researchers found that if they pushed the monitoring goal too hard, the benefits disappeared. When the system was forced to prioritize coverage above all else, the vehicles began to take unnecessarily long detours just to visit specific road blocks. This behavior slowed the fleet down and reduced the overall efficiency of the sensing network. The study identified a "sweet spot" where the fleet was diverse enough to cover the city well but fast enough to remain effective. This balance depended on the size of the fleet; a larger fleet of robotaxis could achieve better coverage with less detouring than a smaller one. The researchers also tested a variation where the system prioritized watching areas with the most cars, rather than just watching every road block equally. This approach successfully increased the number of vehicles observed, though it sometimes required the robotaxis to drive through more congested areas, slightly increasing their travel time.

The study concludes that robotaxi fleets have the potential to serve as a powerful, mobile layer of traffic monitoring, but only if their movement is managed with a specific strategy. Simply letting them drive as they please or forcing them to take long detours does not work. Instead, a coordinated approach that uses the data gathered from monitoring to improve routing decisions can create a positive feedback loop. By turning the fleet into a cooperative sensing network, cities could gain a much richer understanding of traffic dynamics without sacrificing the speed of the service. The researchers suggest that this win-win scenario could provide a strong incentive for robotaxi operators to contribute their vehicles to public monitoring efforts, turning a byproduct of their operation into a valuable resource for the entire city.

Technical Summary: RoboSense – Leveraging Robotaxi Fleets as Drive-by Sensors for Urban Traffic Monitoring

Problem Statement
Urban traffic monitoring is essential for safety analysis, congestion management, and incident response. While traditional infrastructure sensors (e.g., loop detectors) offer temporal continuity, they suffer from limited spatial coverage and high maintenance costs. Conversely, probe vehicles provide spatially continuous trajectories but are constrained by low market penetration rates (MPR) and a lack of control over their routing, as their movements are driven by individual travel needs rather than coordinated data collection. Furthermore, existing probe vehicle data often represents a passive byproduct of travel, lacking the cooperative perception capabilities of Connected and Automated Vehicles (CAVs).

The emergence of robotaxi fleets presents a unique opportunity to address these limitations. Unlike private probe vehicles, robotaxis operate as centrally managed fleets equipped with high-precision onboard perception sensors. They can share real-time data, creating a cooperative perception environment. However, a critical gap exists in dynamically coordinating robotaxi routing to explicitly optimize for traffic monitoring objectives without compromising service efficiency. The core challenge lies in the structural coupling between vehicle mobility and network observability: routing decisions directly determine which parts of the network are observed, yet most existing routing frameworks treat monitoring as an exogenous byproduct rather than an explicit objective.

Methodology
The paper proposes a dynamic routing framework that integrates microscopic simulation, traffic state estimation, and dynamic vehicle routing. The methodology is built upon three primary components:

Cell-Based Network Representation: To align with the sensing resolution of autonomous vehicles (typically 60–120 meters), the urban network is discretized into cells rather than using traditional link-based or path-based representations. This allows the model to capture partial link observations. The Cell Transmission Model (CTM) is employed to represent traffic dynamics, with cell lengths matching the robotaxi detection range (e.g., 80 meters). The CTM predicts traffic states (density, speed) and generates time-dependent travel costs for the routing algorithm.
Spatiotemporal Monitoring Metric: A new performance metric is introduced to quantify monitoring effectiveness. Instead of relying solely on travel efficiency, the framework defines network coverage based on the spatiotemporal occupation of cells by robotaxis. A cell is considered "observed" if at least one robotaxi occupies it, providing complete traffic data for that cell and time step.
Mixed-Integer Linear Programming (MILP) Formulation: The routing problem is formulated as a multi-objective MILP. The objective function jointly minimizes total robotaxi travel time and maximizes spatiotemporal network coverage. The model uses a rolling horizon scheme, where routes are re-optimized periodically (every 100 seconds in the case study) based on updated traffic state predictions from the CTM. The formulation includes constraints for flow conservation, vehicle-level cell occupation, and cell-level aggregation to handle overlapping observations.

Key Contributions
The paper makes three specific contributions:

Integrated Optimization Framework: It embeds traffic monitoring performance as an explicit objective within dynamic robotaxi fleet routing, capturing the intrinsic coupling between vehicle mobility and network observability.
Cell-Based Modeling and Metrics: It adopts a cell-based network representation to support vehicle-level sensing, traffic state estimation, and monitoring evaluation simultaneously. It designs a spatiotemporal monitoring metric based on cell-level observability over time.
Empirical Evaluation: It demonstrates the framework in a microscopic simulation environment (SUMO) using a 5×5 urban grid network. The study evaluates performance across three robotaxi MPRs (2%, 5%, and 10%) and a range of objective weight combinations to illustrate how fleet-level routing decisions influence both monitoring effectiveness and travel efficiency.

Results
Numerical experiments yield several key findings:

Enhanced Monitoring Performance: Incorporating spatiotemporal network coverage into the objective function effectively improves traffic monitoring performance. As the weight of the monitoring objective ( $\alpha_2$ ) increases, the fleet diversifies its routes and extends travel distances to cover more cells, significantly increasing network and vehicle coverage.
The "Win-Win" Scenario: Contrary to the intuition that monitoring and mobility are strictly competing objectives, the study finds that with appropriate weighting, both monitoring performance and robotaxi average speed can be improved simultaneously. This occurs because improved network coverage leads to more accurate traffic state predictions (lower CTM Mean Absolute Error), which in turn enables more efficient route selection, reducing travel time without requiring detours.
Non-Linear Sensing Power: The study evaluates "sensing power" (the fraction of road segments observed). Results indicate that sensing power does not monotonically increase with the monitoring weight. Beyond a certain threshold, increasing the weight forces the algorithm to prioritize coverage uniformity over speed, causing average fleet speeds to drop and reducing overall sensing power. Thus, an optimal balance exists between coverage uniformity, fleet speed, and travel time penalties.
Fleet Size Adaptation: The routing strategy adapts to fleet size. At low MPRs, the algorithm relies on extending individual travel distances to gain coverage. As the fleet size increases, the algorithm shifts toward diversifying route combinations, achieving higher coverage with less travel time penalty.

Significance and Claims
The paper claims to be among the earliest studies to explicitly integrate traffic monitoring objectives into dynamic routing for robotaxi fleets. Its significance lies in demonstrating that traffic monitoring can be transformed from a passive byproduct of individual travel into a purposefully managed data-collection effort.

The authors argue that the identified "win-win" situation—where better network monitoring leads to more accurate traffic state prediction and improved mobility—could incentivize robotaxi operators to contribute their vehicles as drive-by sensors. By leveraging the existing perception hardware of robotaxis and coordinating their movements, operators can potentially improve service quality (via better routing) while providing critical data for urban traffic management. The paper concludes that this approach offers a viable pathway for utilizing robotaxi fleets as an effective, cooperative sensing layer for urban traffic monitoring, provided that routing strategies are carefully balanced to avoid the degradation of sensing power caused by excessive detours or reduced speeds.

More like this