Dynamic Configuration of On-Street Parking Spaces using Multi Agent Reinforcement Learning

TL;DR

This work tackles urban traffic caused by on-street parking by proposing a scalable two-layer MARL framework to dynamically configure parking spaces. It introduces lane-level DQN enhanced with LSTM and GAT to capture spatio-temporal traffic patterns, supervised by block-level agents that maintain neighborhood parking supply. Extensive SUMO-based experiments on synthetic and real Melbourne data demonstrate up to 47% reduction in average travel-time loss with minimal walking-distance penalties and notable emission reductions, outperforming static and other RL baselines. The approach shows strong robustness and practical deployment potential through centralized training with decentralized execution and V2I-enabled information dissemination. It advances infrastructure-centric parking management as a proactive tool for congestion mitigation in smart cities.

Abstract

With increased travelling needs more than ever, traffic congestion has become a major concern in most urban areas. Allocating spaces for on-street parking, further hinders traffic flow, by limiting the effective road width available for driving. With the advancement of vehicle-to-infrastructure connectivity technologies, we explore how the impact of on-street parking on traffic congestion could be minimized, by dynamically configuring on-street parking spaces. Towards that end, we formulate dynamic on-street parking space configuration as an optimization problem, and we follow a data driven approach, considering the nature of our problem. Our proposed solution comprises a two-layer multi agent reinforcement learning based framework, which is inherently scalable to large road networks. The lane level agents are responsible for deciding the optimal parking space configuration for each lane, and we introduce a novel Deep Q-learning architecture which effectively utilizes long short term memory networks and graph attention networks to capture the spatio-temporal correlations evident in the given problem. The block level agents control the actions of the lane level agents and maintain a sufficient level of parking around the block. We conduct a set of comprehensive experiments using SUMO, on both synthetic data as well as real-world data from the city of Melbourne. Our experiments show that the proposed framework could reduce the average travel time loss of vehicles significantly, reaching upto 47%, with a negligible increase in the walking distance for parking.

Figures (9)

• Figure 1: The impact of on-street parking on traffic flow through an intersection illustrated for left hand traffic. (a) Vehicle queues are imbalanced due to parked vehicles which limits traffic throughput through the intersection. (b) Vehicle queues are balanced when few parking spaces are cleared. (c) Parked vehicles obstructing the traffic flow even at the green phase. (d) Minimal impact from parked vehicles when congestion is less.
• Figure 2: Subsequent reductions in average travel time loss obtained by clearing parking spaces one by one.
• Figure 3: Overview of the proposed architecture. Lane level RL agents determine the parking configuration for each lane, and block level agents overlook them to maintain sufficient parking availability in the neighbourhood.
• Figure 4: Deep Q-learning network architecture of the lane level agents. The state vector for the Q-value prediction network consists of three main components; linear observational data captured by a fully connected network, the temporal variation of traffic congestion in the lane captured by LSTM blocks, and the variation of traffic congestion of adjacent lane segments captured by LSTM + GAT blocks. The agent selects the action with the highest Q-value for the current state, predicted by the Q-value prediction network.
• Figure 5: Melbourne suburb region used in the experiments covering 15 intersections, 38 lane segments and 3042 parking spaces.