A framework for human-computer interactive street network design based on a multi-stage deep learning approach

Highlights

• Complement the procedural- and learning-based computer-aided frameworks for street network design.

• Improve conventional, end-to-end deep-learning structure to a multi-stage one.

• Develop an HCI system for interactive and progressive street network design.

• Provide more realistic planning options and enrich alternatives in the early stages of urban planning and design.

Abstract

Limited attention has been given to human-computer interactions in the plan-making process to capitalize on the relative strengths of both. This paper proposes a methodological framework for an interactive street network design that complements user-driven (i.e., procedural-based tools) and example-driven (i.e., learning-based tools) approaches in urban planning and design. The proposed framework consists of three components: (1) a data preparation module to link open-source road networks with human-labeled planning guidance, (2) a multi-stage deep learning (MSDL) model to reinforce the user-defined guidance in the automatic generation of street networks, and (3) a human-computer interaction (HCI) interface to enable the progressive design process. The performance and the working mechanism of the proposed framework were examined through experiments in four European cities (i.e., Amsterdam, Barcelona, Berlin, and Prague). The experiments demonstrate that the proposed MSDL model can achieve a better predictive performance compared to benchmark models, particularly when limited planning guidance is given. These finding are revealed using either computer vision- or street network-related metrics. With less than 40% of the ground truth planning guidance used as an input, the MSDL model can perform as well as other models using 100% of the information. Furthermore, when embedded within an HCI system for user trials, the model can facilitate a human-computer collaborative design process. This advantage is derived from the model's ability to provide initial prototypes, timely responses to changed guidance, and quantifiable evaluations of the generated proposals. Suitable for professionals and laypersons, the proposed tool can inform plan-making and public engagement by offering realistic, enriched, and human-centered spatial proposal alternatives for comparison.

Introduction

The accelerated urban expansion and regeneration that is driven by the increasing world population narrow the windows of time for urban planning and design (UN-Habitat, 2020). These increasing demands lead to the advancement of computer-aided planning and design tools used in the automatic or guided generation of spatial proposals, particularly in the preliminary stages of urban (re)development (Brömmelstroet, 2013; Deal et al., 2017; Yang et al., 2019). Among the various functions of computer-aided tools, street network planning and design is regarded as the backbone of generating urban spatial layouts and shaping the built environment.

A series of methods have been developed to support the planning and design of street configurations. For the conventional drawing-based design process, the concepts of “link and place” and “cooperative design” have been widely adopted to guide proposal development, which considers transport connectivity, street activities, and stakeholders' feedback (see, e.g., Jones, Marshall, & Boujenko, 2008, Laurini, 2001). For computer-based decision support tools, either rule sets or precedents are manually defined or fed into the computer systems to realize the automatic generation of urban street layouts (see, e.g., Parish & Müller, 2001).

However, user-driven design (e.g., drawing-based and procedural/rule-based approaches) overly relies on planners' disciplinary knowledge and professional experience in the quality delivery of the design proposals. The design process is led by professionals whose seniority plays a decisive role in coordinating the stakeholders' engagement. By contrast, example-driven design (e.g., learning-based approaches) focuses on the automatic generation of street configurations based on precedents, with limited consideration of users' intelligence and guidance.

To date, little attention has been given to the interactions between human and machine intelligence in the street network design process to complement the strengths of both. Among the limited relevant attempts, Fang, Jin, and Yang (2021) explored the potential incorporation of users' knowledge (e.g., design guidance) into computer models (e.g., deep learning methods) for delivering spatial proposals that can better fit within the surrounding context. Nevertheless, the method remains outcome-oriented (“click and go”), where the dynamic processes of the proposal development (i.e., the human-computer interactions) cannot be achieved.

Humans are adept at local knowledge and professional intelligence, while computers offer automatic generation and machine intelligence. To capitalize on these relative strengths for urban planning and design, this research develops a methodological framework for interactive street network design based on a multi-stage deep learning approach. The framework consists of three modules: (1) a dataset preparation module that links open-source road networks with human-labeled planning guidance; (2) a deep learning module for the automatic generation of urban road networks. The module extends the conventional end-to-end structure of deep learning techniques in street network generation (see, e.g., Fang et al., 2021) to a multi-stage model, thereby enabling the reinforcement of user-defined and iteratively developed guidance systems; and (3) a human-computer interaction (HCI) system that can facilitate the interactive and progressive design process.

The proposed framework is tested in a case study of four European cities—Amsterdam, Barcelona, Berlin, and Prague. The four cities have representative urban (re)development processes for street configurations that date from the medieval period, characterized by a rich mixture of street network patterns. Note that although the proposed framework can be universally applied to different case study areas, users are expected to follow the dataset preparation module to develop a localized dataset for model training due to varying factors that determine local street network design globally, such as history, culture, and topography.

The remainder of this paper is organized as follows. Section 2 presents a literature review on computer-aided tools for street network design. Section 3 illustrates the proposed framework for interactive street network generation. Section 4 discusses the datasets for the model training and the experiment design used for evaluating the models' performance. Section 5 presents the results of the experiment centered around the MSDL model and its ability to work within the HCI framework. Section 6 concludes the paper and discusses its contribution.