Automated Architectural Space Composition for Building Renovation Using Deep Reinforcement Learning

The increasing demand for old building renovation presents complex spatial, functional, and technical challenges that exceed the capacity of conventional design workflows. In response, this paper investigates the application of deep reinforcement learning (DRL) to enable automated architectural space composition under predefined built environment constraints. By integrating artificial intelligence into architectural design processes, the study positions computational intelligence as a strategic tool to improve efficiency, adaptability, and decision-making in renovation-oriented architectural practice.

Reinforcement Learning Framework for Architectural Design

The research establishes a dedicated reinforcement learning platform centered on the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm. This framework enables multiple agents to collaboratively explore spatial configurations while responding to shared environmental constraints. By framing architectural space composition as a sequential decision-making problem, the study demonstrates how reinforcement learning can effectively simulate design logic, spatial negotiation, and constraint satisfaction in renovation scenarios.

Functional Blocks as Agents in Parametric Modeling

Within the Grasshopper parametric environment, functional blocks are defined as intelligent agents representing architectural components and spatial units. These agents interact dynamically, simulating existing building structures and their spatial relationships. This section discusses how agent-based modeling allows architectural elements to respond autonomously to environmental conditions and design objectives, enabling a flexible and scalable approach to spatial reconfiguration in renovation projects.

Task-Based Reward System and Design Guidance

A task-oriented reward mechanism is developed to guide the learning and decision-making process of the agents. Rewards are assigned based on spatial performance, functional suitability, and compliance with renovation objectives. This topic explains how the reward system translates architectural intent into quantifiable learning signals, effectively steering automated design exploration toward viable and optimized space compositions.

Algorithm–Model Integration and Training Performance

To ensure seamless interaction between the reinforcement learning algorithm and the architectural modeling platform, data exchange is achieved through UDP communication. Training experiments conducted on typical frame-structure spaces reveal a substantial increase in cumulative rewards around the 650,000th training step. This section analyzes the training outcomes, demonstrating the system’s ability to converge toward effective spatial solutions and successfully accomplish predefined design tasks.

Implications for Human–Machine Collaborative Renovation Design

The findings highlight the potential of deep reinforcement learning–based automated space composition to significantly enhance design efficiency and spatial optimization in building renovation projects. Rather than replacing architects, the proposed approach supports human–machine collaboration, where designers guide objectives and constraints while intelligent systems explore and optimize solutions. This research contributes to the evolving discourse on computational design methodologies and their role in shaping the future of adaptive, intelligent architectural practice.

Architecture Engineers Awards

🔗 Nominate now! 👉 https://architectureengineers.com/award-nomination/?ecategory=Awards&rcategory=Awardee 🌐 Visit: architectureengineers.com 📩 Contact: contact@architectureengineers.com Get Connected Here: ***************** Instagram :  https://www.instagram.com/architecture_engineers_awards/ Facebook :  https://www.facebook.com/profile.php?id=61576995475934 Tumblr :   https://www.tumblr.com/architectureengineersawards Pinterest :   https://in.pinterest.com/architectureengineersawards/ Blogger :   https://architectureengineers.blogspot.com/ Twitter :   https://x.com/Architectu54920 YouTube :  https://www.youtube.com/@Architechtureengineer LinkedIn :  https://www.linkedin.com/in/architecture-engineer-01a044361/

#ArchitecturalAutomation
#DeepReinforcementLearning
#AutomatedDesign
#BuildingRenovation
#ComputationalArchitecture
#MultiAgentSystems
#MADDPG
#ParametricDesign
#Grasshopper3D
#SpatialOptimization
#AIinArchitecture
#DesignTechnology
#HumanMachineCollaboration
#SmartDesignSystems
#ArchitecturalResearch
#DigitalDesignMethods
#IntelligentBuildings
#DesignEfficiency
#AdaptiveArchitecture
#FutureOfArchitecture