Evolutionary City: Towards a Flexible, Agile and Symbiotic System

Ding Wang, Xi Chen, Wei Hu, Jingru Yu, Ding Wang 0001, Shengyue Yao, Yilun Lin, Fei-Yue WangarXiv 2023 · DOI · arXiv

Abstract

Urban infrastructure has traditionally been rigid, struggling to keep pace with the volatile demands of modern "City 5.0" environments. This paper proposes the "Evolutionary City" framework, which leverages real-time multi-modal sensing, Graph Neural Network (GNN) simulations, and blockchain-based decentralized autonomous organizations (DAOs) to create a symbiotic urban ecosystem. By using Large Language Models (LLMs) as an interface for human-AI alignment, the system can dynamically reconfigure resources like traffic lanes. Experimental results from a case study in Shanghai show significant improvements in traffic efficiency (17.46%) and environmental impact (20.53% reduction in emissions).

Introduction: The Challenge of Rigid Infrastructure

Traditional urban infrastructure is inherently "rigid." Once a road is constructed or a traffic pattern established, modification is both time-consuming and capital-intensive. In the emerging "City 5.0" era, however, urban demand is increasingly volatile, driven by AI-integrated equipment and unpredictable system-wide emergencies.

The supply-demand gap in rigid urban systems.

The core issue is the "supply-demand gap": physical infrastructure (supply) cannot evolve at the same speed as resident behavior (demand). Current management paradigms suffer from high latency in perception, low prediction accuracy, and a lack of trust between human administrators and autonomous AI systems.

The Evolutionary City Framework

To address these challenges, the authors introduce a four-pillar framework designed to transform the city into a "living organism" that mutates its management strategies based on environmental feedback.

The four-pillar Symbiotic Intelligence Mechanism.
  1. Adaptability Perception: This layer uses a network of "nerve endings"—combining official sensors with resident-owned devices—to monitor urban states in real-time. Privacy-Enhancing Technologies (PET) are integrated to ensure data security and encourage public participation.
  2. Parallel Simulation: Utilizing Graph Neural Networks (GNNs), the framework encodes city components as nodes and edges. This allows for multi-scale simulations that predict the outcomes of various "evolutionary trajectories" in a virtual space before they are deployed.
  3. Autonomous Decision-Making (DAO): Decentralized Autonomous Organizations (DAOs) on the blockchain manage complex interactions between intelligent agents (e.g., traffic signals), ensuring consensus and eliminating single points of failure.
  4. Symbiotic Intelligence (LLM Interface): A novel bridge using Large Language Models (LLMs) allows human operators to set high-level goals in natural language. The LLM translates these intents into mathematical constraints for optimization algorithms using a hybrid AHP-EWM weighting method.

Case Study: Shanghai Linkong SOHO

The framework was validated using a real-world case study in the Shanghai Linkong SOHO Economic Park, focusing on the management of variable lane clusters. The study utilized real-time traffic flow data from nine major access roads during peak hours. Simulations were conducted using both rule-based (VISSIM) and GNN-based (TransworldNG) tools to cross-validate the evolutionary strategies.

Key Results and Insights

The implementation of "soft" infrastructure—specifically reversible lanes that change direction based on tidal traffic—yielded significant gains:

  • Traffic Efficiency: Increased by 17.46%.
  • Environmental Impact: Emissions (CO2, NOx) and fuel consumption were reduced by 20.53%.
  • Travel Delay: Practical testing showed a reduction in delays between 15% and 20%.

The system proved most effective during morning peaks, where the supply-demand mismatch is typically most severe. By using LLMs to translate human values into machine-executable constraints, the framework ensures that AI optimization remains aligned with societal needs.

Conclusion and Future Work

The "Evolutionary City" represents a shift toward agile, trust-based urban management. While successful in a localized economic park, future work must address scalability to city-wide levels and refine human-AI conflict resolution. The ultimate goal is a "lane-free" automated traffic system where the city's capacity evolves seamlessly in real-time.

Summary updated

BibTeX
@misc{wang2023evolutionary,
  author = {Ding Wang and Xi Chen and Wei Hu and Jingru Yu and Ding Wang and Shengyue Yao and Yilun Lin and Fei-Yue Wang},
  title = {{Evolutionary City: Towards a Flexible, Agile and Symbiotic System}},
  year = {2023},
  eprint = {2311.14690},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2311.14690},
}
Figure