Proposal for a Framework for Optimizing Artificial Environments Based on Physiological Feedback

Hideyuki Takagi, Shangfei Wang, Shota NakanoJournal of PHYSIOLOGICAL ANTHROPOLOGY and Applied Human Science 2005 · DOI

Abstract

We propose and then evaluate a new framework for finding the physical parameters of an artificial environment which give rise to given target physiological characteristics. We assume that a human is a system that takes as inputs the physical parameters of an artificial environment and outputs physiological parameters in response. We define our task as the inverse problem; we must find the best inputs from given target outputs. Our proposed framework solves the inverse problem using evolutionary computation techniques to optimize an artificial environment. We evaluate this framework using a simulation with a vibration environment and verify that it works.

Introduction

Modern life is largely spent within artificial environments where factors like temperature, lighting, and sound are tightly controlled. Traditionally, environmental design follows a "feed-forward" approach: researchers adjust environmental settings and observe the resulting human physiological response. However, a significant challenge arises when we have a specific physiological goal—such as achieving a target heart rate or blood pressure—and need to determine the precise environmental parameters to elicit that state.

This is an "inverse problem." Because the human body is a complex, non-linear system, working backward from a desired output to the necessary inputs is computationally difficult and often relies on inefficient trial-and-error. This paper proposes a data-driven framework to automate this optimization process using evolutionary computation.

The Proposed Framework

The core of the proposal is a Physiological Feedback-based Interactive Evolutionary Computation (IEC) framework. Unlike traditional IEC, which relies on subjective human "likes" or "dislikes" to guide evolution, this system uses objective physiological data to drive the optimization.

The proposed framework for optimizing artificial environments using physiological feedback.

The framework operates in a continuous loop:

  1. Genetic Algorithm (GA): Generates a population of environmental parameter sets (e.g., vibration frequencies and amplitudes).
  2. Measurement: The human subject (or a simulation) is exposed to these parameters, and their physiological response is measured.
  3. Fitness Evaluation: The system calculates the "fitness" of each parameter set based on the Euclidean distance between the measured physiological state and the predefined target state.
  4. Evolution: The GA selects the best-performing parameters to create the next generation, iteratively narrowing the gap between the current and target states.

Experimental Simulation

To validate the framework without subjecting humans to exhausting, repetitive testing, the researchers employed "virtual humans." They first collected data from 15 subjects exposed to 12 different vibrational conditions in a specialized chair. This data was used to train 15 individual Neural Networks (NNs), each representing the unique physiological response profile of a specific person.

The optimization task involved searching an 8-dimensional vibration space (including axes like pitch, roll, and yaw) to reach a target 4-dimensional physiological vector consisting of systolic/diastolic blood pressure, heart rate, and breathing frequency.

Key Results and Discussion

The simulation results confirmed the framework's effectiveness. Across all 15 virtual subjects and 16 different target points, the Genetic Algorithm successfully converged toward the desired physiological states.

Convergence of the genetic algorithm across 20 generations for different subjects.

As shown in the figure above, the "distance" to the target state consistently decreased over 20 generations. A critical insight from the study was the sheer difficulty of the search: the researchers estimated that the "target area" for a desired physiological state occupied only 0.016% of the total search space. This highlights why manual adjustment is so often unsuccessful in complex environments.

While the system converged, the authors noted that the final proximity to the global optimum was limited by the small population size (10) and generation count (20). These constraints were intentionally chosen to reflect the practical limits of real-world human testing, where fatigue would prevent longer experiments.

Conclusion and Future Work

This research demonstrates that evolutionary computation can effectively solve the inverse problem of environmental design by using physiological feedback. By moving from subjective preferences to objective sensor data, we can create "silent" optimization systems that tailor environments to specific health or comfort goals.

Future work will focus on validating this framework with real-time measurements on human subjects. Additionally, addressing data scarcity and the high variability between individuals will be crucial for scaling this technology to broader applications in smart homes, vehicles, and workspaces.

Summary updated

Abstract

We propose and then evaluate a new framework for finding the physical parameters of an artificial environment which give rise to given target physiological characteristics. We assume that a human is a system that takes as inputs the physical parameters of an artificial environment and outputs physiological parameters in response. We define our task as the inverse problem; we must find the best inputs from given target outputs. Our proposed framework solves the inverse problem using evolutionary computation techniques to optimize an artificial environment. We evaluate this framework using a simulation with a vibration environment and verify that it works.
BibTeX
@article{takagi2005proposal,
  author = {Hideyuki Takagi and Shangfei Wang and Shota Nakano},
  title = {{Proposal for a Framework for Optimizing Artificial Environments Based on Physiological Feedback}},
  journal = {Journal of PHYSIOLOGICAL ANTHROPOLOGY and Applied Human Science},
  year = {2005},
  doi = {10.2114/jpa.24.77},
}
Figure