This thesis investigates the smart home as a historically evolving “interior-as-interface,” tracing a genealogy from early modernist rationalization to post-digital, multimodal responsiveness, and culminates in a design-led, low-barrier planning agent for non-technical residents. Rather than a linear gadget history, the work frames domestic space as a shifting cultural apparatus in which modernization, media, and control progressively reconfigure relations among inhabitants, environments, and devices, recasting the home as an intelligible, responsive medium for orchestrated scenarios. Building on a theoretical lens where agency is distributed across sensing, reasoning, and action, the thesis positions multimodal AI as an architectural actor that must remain legible, interruptible, and trustworthy. Empirically, a survey maps today’s adoption plateau and pain points: households typically own 3–5 devices; fragmentation, reliability, and complexity dominate barriers; and users express a preference for local processing, explicit consent, and revocable authorization. Responding to these findings, the project prototypes a two-part AI system—Retrieval-Augmented Q&A and a Personalized Planning Assistant—implemented as orchestrated Coze agents and grounded in a Xiaomi/Mijia knowledge base for verifiable plans. The pipeline collects preferences and space information, extracts pain points, and composes room-level device lists with Condition→Action automations checked against capability tables. The contribution is twofold: a historically/theoretically grounded account of the domestic interior’s transition to a multimodal interface, and a practical, human-centered agent that lowers setup complexity while keeping control perceptible and reversible. The thesis concludes with limitations and a roadmap toward broader ecosystems (e.g., Matter) and stronger spatial understanding—bridging designer intent and user reality for smart homes that truly serve all residents.
Questa tesi indaga la smart home come “interno-come-interfaccia” in evoluzione storica, tracciandone una genealogia dalla razionalizzazione modernista alla reattività post-digitale e multimodale, e culminando in un agente di pianificazione guidato dal design e a bassa soglia per residenti non tecnici. Più che una storia lineare dei gadget, il lavoro inquadra lo spazio domestico come un dispositivo culturale mutevole in cui modernizzazione, media e controllo riconfigurano progressivamente le relazioni tra abitanti, ambienti e dispositivi, ridefinendo la casa come un medium intelligibile e responsivo per scenari orchestrati. Su una cornice teorica in cui l’agency è distribuita tra percezione, ragionamento e azione, la tesi posiziona l’IA multimodale come attore architettonico che deve rimanere leggibile, interrompibile e affidabile. Empiricamente, un sondaggio mappa l’attuale plateau di adozione e i punti dolenti: le famiglie possiedono tipicamente 3–5 dispositivi; frammentazione, affidabilità e complessità rappresentano le principali barriere; gli utenti esprimono preferenza per l’elaborazione locale, il consenso esplicito e autorizzazioni revocabili. In risposta, il progetto prototipa un sistema di IA in due parti—Q&A con recupero aumentato e un Assistente di Pianificazione Personalizzato—implementato come agenti Coze orchestrati e ancorato a una knowledge base Xiaomi/Mijia per piani verificabili. La pipeline raccoglie preferenze e dati spaziali, estrae i pain point e compone liste di dispositivi per stanza con automazioni Condizione→Azione verificate rispetto a tabelle di capacità. Il contributo è duplice: un resoconto storico-teorico della transizione dell’interno domestico verso un’interfaccia multimodale e un agente pratico, centrato sulla persona, che riduce la complessità di setup mantenendo il controllo percepibile e reversibile. La tesi si conclude con i limiti e una roadmap verso ecosistemi più ampi (ad es. Matter) e una comprensione spaziale più solida—ponendo un ponte tra intento del progettista e realtà d’uso per smart home che servano davvero tutti gli abitanti.
Smart home, simple setup: designing a low-barrier AI agent for room-by-room planning and control
Li, Jiahang
2025/2026
Abstract
This thesis investigates the smart home as a historically evolving “interior-as-interface,” tracing a genealogy from early modernist rationalization to post-digital, multimodal responsiveness, and culminates in a design-led, low-barrier planning agent for non-technical residents. Rather than a linear gadget history, the work frames domestic space as a shifting cultural apparatus in which modernization, media, and control progressively reconfigure relations among inhabitants, environments, and devices, recasting the home as an intelligible, responsive medium for orchestrated scenarios. Building on a theoretical lens where agency is distributed across sensing, reasoning, and action, the thesis positions multimodal AI as an architectural actor that must remain legible, interruptible, and trustworthy. Empirically, a survey maps today’s adoption plateau and pain points: households typically own 3–5 devices; fragmentation, reliability, and complexity dominate barriers; and users express a preference for local processing, explicit consent, and revocable authorization. Responding to these findings, the project prototypes a two-part AI system—Retrieval-Augmented Q&A and a Personalized Planning Assistant—implemented as orchestrated Coze agents and grounded in a Xiaomi/Mijia knowledge base for verifiable plans. The pipeline collects preferences and space information, extracts pain points, and composes room-level device lists with Condition→Action automations checked against capability tables. The contribution is twofold: a historically/theoretically grounded account of the domestic interior’s transition to a multimodal interface, and a practical, human-centered agent that lowers setup complexity while keeping control perceptible and reversible. The thesis concludes with limitations and a roadmap toward broader ecosystems (e.g., Matter) and stronger spatial understanding—bridging designer intent and user reality for smart homes that truly serve all residents.| File | Dimensione | Formato | |
|---|---|---|---|
|
2025_12_Li_Tesi.pdf
accessibile in internet per tutti
Dimensione
18.77 MB
Formato
Adobe PDF
|
18.77 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/246077