A computational design thesis on emotion-aware interaction.
THE MAZE began as a Master's thesis in 2021 and has been rebuilt as a prototype that explores how reinforcement learning, applied to visible affective response, can quietly inform design decisions in colour and architecture.
Origin
THE MAZE began as a Master's thesis in computational design at Technische Hochschule Ostwestfalen-Lippe in Detmold, Germany, in 2021 — an attempt to surface a quiet conversation between a viewer and a visual environment. The premise is simple: rather than asking what someone likes, observe how they respond.
From thesis to prototype
The thesis explored the principle. This prototype rebuilds it on the present toolset — a modern browser stack, on-device affect estimation, hybrid reinforcement learning, and a calmer interface — and adds the design preference review module on top of the original colour work.
Approach
We treat visible response — attention, calm, engagement, discomfort — as a stream of soft signals. None of these signals are diagnostic on their own. Together, across many short trials, they form a noisy preference function the system tries to learn.
Reinforcement learning
A lightweight contextual bandit selects what to show next. Early in a session it explores broadly across the spectrum or the design set. As confidence grows, a tabular Q-learner and a small DQN refine the policy, narrowing toward regions of strongest response.
Application
Two near-term applications. The first is personal — a calm self-knowledge tool. The second is professional — a quiet instrument for architects and interior designers presenting options to a client, generating a defensible, response-based report.
Ethics & framing
Visible response is not a window into the mind. It is a measurement of behaviour. THE MAZE does not diagnose, identify, or store identity. It produces estimates, framed as estimates, and is meant to support — never replace — design judgement.