Balancing Mobility Efficiency and Environmental Exposure During Mega Events: A Multi-agent Routing Framework

Xinyue Liu; Qiumeng Li; Suhong Zhou; Junwen Lu; Jiali Zhou; Weipan Xu

cs.MAOct 26, 2026

Balancing Mobility Efficiency and Environmental Exposure During Mega Events: A Multi-agent Routing Framework

Authors: Xinyue Liu, Qiumeng Li, Suhong Zhou, Junwen Lu, Jiali Zhou, Weipan Xu

Organizations: Thrust of Urban Governance and Design, Society Hub, The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, 511453, China · Thrust of Intelligent Transportation, Systems Hub, The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, 511453, China · School of Geography and Planning, Sun Yat-sen University, Guangzhou, 510006, China · School of Intelligent Systems Engineering, Sun Yat-sen University, Guangzhou, 510006, China ·

Abstract

Mega events can generate concentrated travel demand, intensifying congestion around venues and increasing travelers' environmental exposure. Existing traffic management strategies focus on travel efficiency, with limited attention to exposure risks. Taking mega-event travel as a short-term high-demand scenario, this study develops an exposure-aware multi-agent routing framework integrating road-network conditions, event-induced demand, background traffic, and environmental data within a unified simulation environment. Motorized travelers optimize travel efficiency, whereas micro-mobility travelers consider travel time and dynamic exposure to traffic-related PM2.5 and noise. We compare the framework with shortest-distance, congestion-aware, and exposure-aware routing baselines. Results show that, under high demand, the SARSA-based framework reduces vehicle travel time by 34.59% and cumulative environmental exposure by 58.65%, while micro-mobility travel time remains nearly unchanged. Exposure reduction is jointly driven by vehicle-side congestion mitigation and low-exposure routing. The framework provides quantitative support for vehicle diversion, low-exposure route guidance, and health-oriented traffic management in short-term high-demand scenarios.

Figures & tables

Method	Vehicle routing	Micro-mobility routing
Shortest-distance	Shortest road-graph distance	Shortest road-graph distance
Congestion-aware	Current congestion-dependent travel time
Static exposure-aware		Fixed background exposure-weighted cost
Dynamic exposure-aware		Current exposure-weighted cost; replanned after each crossing
Minimum-exposure		Fixed background exposure-priority cost

Table 1: Nonlearning baseline routing strategies.

Figure 2: SARSA training rewards on the 250 m grid. Lines and shaded bands represent five-seed means and 95% confidence intervals, respectively.

Figure 3: Routing-strategy performance on the 250 m grid, including vehicle travel time, micro-mobility travel time, and environmental exposure. Points represent five-seed means, with five matched tests per seed.

$N$	Method	$T_{\mathrm{veh}}$ mean [95% CI]	$\Delta T_{\mathrm{veh}}$ (%)	$T_{\mathrm{mic}}$ mean [95% CI]	$\Delta T_{\mathrm{mic}}$ (%)	$EI^{\Delta}$ mean [95% CI]	$\Delta EI^{\Delta}$ (%)
40,000	Shortest-distance	76.0 [75.8, 76.2]	–	1173.4 [1173.4, 1173.4]	–	2.6 [2.6, 2.6]	–
	Congestion-aware	68.1 [68.1, 68.1]	-10.33	1173.4 [1173.4, 1173.4]	+0.00	1.6 [1.6, 1.6]	-36.98
	Static exposure-aware	68.1 [68.1, 68.1]	-10.32	1188.4 [1188.4, 1188.4]	+1.28	1.4 [1.4, 1.4]	-45.33
	Dynamic exposure-aware	68.1 [68.1, 68.1]	-10.33	1173.4 [1173.4, 1173.4]	+0.00	1.5 [1.5, 1.5]	-43.77
	Minimum-exposure	68.1 [68.1, 68.1]	-10.32	1717.0 [1717.0, 1717.0]	+46.34	0.9 [0.9, 0.9]	-65.92
	MAPPO	71.5 [68.6, 74.4]	-5.89	1173.4 [1173.4, 1173.4]	+0.00	1.5 [1.4, 1.6]	-43.59

Table 2: Routing-strategy performance on the 250 m grid. Values are five-seed means [95% confidence intervals], with five tests per seed. Time:

\times 10^{3}

min; exposure:

\times 10^{3}

index-min. Changes are relative to shortest-distance routing; totals include incomplete trips.

Figure 4: Spatial patterns from the first frozen-policy test at $N=80{,}000$ for seed 42 under shortest-distance routing, dynamic exposure-aware routing, and SARSA. Columns show vehicle traffic load, micro-mobility traffic load, and congestion-induced exposure.

$\eta$	$T_{\mathrm{mic}}$ mean [95% CI]	$\Delta T_{\mathrm{mic}}$ mean (%)	$EI^{\Delta}$ mean [95% CI]	$\Delta EI^{\Delta}$ mean (%)
0.0	2341.8 [2341.8, 2341.8]	–	6.5 [6.2, 6.7]	–
1.0	2341.8 [2341.8, 2341.8]	+0.00	3.9 [3.9, 4.0]	-38.77
3.0	2341.8 [2341.8, 2341.8]	+0.00	3.9 [3.8, 4.0]	-39.73
5.0	2341.8 [2341.8, 2341.8]	+0.00	3.9 [3.8, 4.0]	-39.77
7.0	2341.8 [2341.8, 2341.8]	+0.00	3.8 [3.8, 3.9]	-40.48

Table 3: SARSA exposure-weight sensitivity at

N=80{,}000

on the 250 m grid. Values are five-seed means [95% confidence intervals], with five tests per seed. Time:

\times 10^{3}

min; exposure:

\times 10^{3}

index-min. Changes are relative to

\eta=0

Target obedience (%)	Realized obedience vehicle/micro (%)	$T_{\mathrm{veh}}$ mean [95% CI]	$T_{\mathrm{mic}}$ mean [95% CI]	$EI^{\Delta}$ mean [95% CI]	$\Delta EI^{\Delta}$ mean (%)
0	0.00/0.00	151.8 [151.8, 151.8]	2341.8 [2341.8, 2341.8]	6.6 [6.6, 6.6]	-29.49
25	25.08/25.02	143.2 [142.5, 143.8]	2341.8 [2341.8, 2341.8]	6.4 [6.4, 6.5]	-31.47
50	50.10/49.98	138.3 [137.5, 139.0]	2341.8 [2341.8, 2341.8]	6.0 [5.9, 6.1]	-36.32
75	75.21/74.98	136.6 [136.4, 136.8]	2341.8 [2341.8, 2341.8]	5.3 [5.2, 5.3]	-43.79
100	100.00/100.00	136.7 [136.7, 136.8]	2341.8 [2341.8, 2341.8]	3.9 [3.8, 4.0]	-58.65

Table 4: SARSA obedience sensitivity at

N=80{,}000

on the 250 m grid. Values are five-seed means [95% confidence intervals], with five tests per seed. Time:

\times 10^{3}

min; exposure:

\times 10^{3}

index-min. Exposure changes are relative to shortest-distance routing; realized rates denote vehicle/micro-mobility obedience.

Appendix figures & tables5 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 6: MAPPO training rewards on the 250 m grid. Lines and shaded bands show five-seed means and 95% confidence intervals.

	250 m	Multiscale
Coefficient	Main / exposure-weight	100 m	150 m	200 m
Vehicle time $w_{\mathrm{veh}}$	1	5	5	5
Micro-mobility time $w_{\mathrm{mic}}$	0.005	0.005	0.005	0.005
Exposure $\eta$	$3^{\mathrm{a}}$	3	3	3
Vehicle progress $w^{\mathrm{veh}}_{p}$	0.1	2.5	$10/3$	1.25
Additional vehicle progress $\beta$	0	0	$500/36$	0

Appendix

Table 5: SARSA reward coefficients.

Grid (m)	Method	$T_{\mathrm{veh}}$ mean [95% CI]	$\Delta T_{\mathrm{veh}}$ (%)	$T_{\mathrm{mic}}$ mean [95% CI]	$\Delta T_{\mathrm{mic}}$ (%)	$EI^{\Delta}$ mean [95% CI]	$\Delta EI^{\Delta}$ (%)
100	Shortest-distance	470.2 [468.7, 471.7]	–	2762.7 [2761.7, 2763.7]	–	34.2 [34.2, 34.3]	–
	Congestion-aware	402.6 [401.8, 403.3]	-14.38	2762.7 [2761.7, 2763.7]	+0.00	30.8 [30.6, 31.0]	-10.18
	Static exposure-aware	404.0 [402.3, 405.8]	-14.07	2782.3 [2781.2, 2783.4]	+0.71	19.6 [19.5, 19.7]	-42.77
	Dynamic exposure-aware	402.6 [401.8, 403.3]	-14.38	2762.9 [2761.9, 2763.9]	+0.00	22.3 [22.1, 22.5]	-34.83
	Minimum-exposure	403.9 [402.3, 405.6]	-14.09	4105.4 [4102.2, 4108.6]	+48.60	8.7 [8.7, 8.7]	-74.67
	SARSA	335.5 [321.9, 349.1]	-28.65	2762.7 [2761.7, 2763.7]	+0.00	12.4 [10.5, 14.3]	-63.82

Appendix

Table 6: Grid-resolution sensitivity at

N=80{,}000

. Values are five-seed means [95% confidence intervals], with five matched tests per seed. Time:

\times 10^{3}

min; exposure:

\times 10^{3}

index-min. Changes are relative to shortest-distance routing; totals include incomplete trips.

Figure 7: Test spatial patterns for the four additional routing strategies, using the same columns, conditions, and scales as Figure 4 .

Risk-Aware Path Planning

See all 23 papers

Decentralized Multi-Agent Urban Traffic Management via Spatio-Temporal Mobility Profile Planning
Aug 8, 2026Lorenzo Mario Amorosa, Lorenzo Farina, Vittorio Todisco +1Multi-Agent Coordination Intelligent Transportation Systems
Department of Electrical, Electronic and Information Engineering (DEI), “Guglielmo Marconi”, University of Bologna & WiLab - National Wireless Communications Laboratory of CNIT, Bologna, Italy.
Dynamic Deployment of Mobile Charging Trucks During Natural Disaster Evacuation: An Offline-to-Online Framework
May 16, 2026Rui Ma, Zilin Bian, Kaan OzbayElectric Vehicle Charging Vehicle Routing
New York University · Rochester Institute of Technology
Learning to Incentivise: Using Reinforcement Learning for Sustainable Urban Mobility
Oct 21, 2026Germán Pardo-González, Shaghayegh Vosough, Katerina Papadaki +1Intelligent Transportation Systems
Dept. of Mathematics, London School of Economics and Political Science, London, UK · Dept. of Built Environment, Aalto University, Espoo, Finland · Centre for Industrial Management/Traffic and Infrastructure, KU Leuven, Leuven, Belgium

Aug 8, 2026J/K move · Enter open · S save

Decentralized Multi-Agent Urban Traffic Management via Spatio-Temporal Mobility Profile Planning

Lorenzo Mario Amorosa, Lorenzo Farina, Vittorio Todisco, Alessandro Bazzi

Department of Electrical, Electronic and Information Engineering (DEI), “Guglielmo Marconi”, University of Bologna & WiLab - National Wireless Communications Laboratory of CNIT, Bologna, Italy.

Multi-Agent Coordination Intelligent Transportation Systems Autonomous Driving Planning Multi-Agent Path Finding

Open paper AskPDF arXiv

As modern cities face increasingly severe traffic congestion, connected and autonomous vehicles (CAVs) have emerged as a crucial enabling technology for next-generation intelligent traffic management. However, fully realizing this potential is hindered by the limitations of current paradigms. Existing approaches typically optimize localized interactions rather than system-wide efficiency, incur severe communication overhead, or lack the deterministic guarantees required for safe kinematic execution. Furthermore, current multi-agent adaptations are frequently restricted to small predefined scenarios, failing to scale across large and complex urban networks. To bridge this gap, this paper introduces VeloCity, a decentralized multi-agent spatio-temporal mobility profile planning framework designed for CAVs operating in arbitrary urban areas. To minimize vehicles' travel times, VeloCity distributes mobility profile optimization directly to individual CAVs. Vehicles query a localized traffic coordinator for a reservation table, independently compute their fastest conflict-free mobility profile, and reserve their requested space-time slots back with the coordinator. By natively adapting to any arbitrary road topology, the framework manages highly irregular urban areas without requiring scenario-specific tuning, all while guaranteeing collision-free and physically executable vehicle trajectories. Extensive simulations across four large-scale real-world urban maps (Tokyo, Manhattan, Rome, and Bologna) demonstrate the framework's scalability. Compared to established state-of-the-art models, VeloCity yields drastically lower travel times, tightly bounds delay variance, and successfully prevents congestion gridlocks even under extremely high vehicular densities.

Abstract

Risk-Aware Path Planning

Explore similar work

Decentralized Multi-Agent Urban Traffic Management via Spatio-Temporal Mobility Profile Planning

Dynamic Deployment of Mobile Charging Trucks During Natural Disaster Evacuation: An Offline-to-Online Framework

Learning to Incentivise: Using Reinforcement Learning for Sustainable Urban Mobility