From chopped to cooked: Design and inference in physical environments

Link to paper

@inproceedings{yang2026environmentdesign,
  title = {From chopped to cooked: Design and inference in physical environments},
  booktitle = {Proceedings of the 48th {Annual} {Conference} of the {Cognitive} {Science} {Society}},
  author = {Yang, Justin and Wong, Lionel and Fan, Judith E. and Gerstenberg, Tobias},
  year = {2026},
}

Contents:

• Overview
• Repository structure
• Set up
• Experiments
• Empirical analyses
• External resources
• CRediT author statement

Overview

Physical spaces are often designed to support specific uses. But how do people create such environments, and how do users infer their intended function? We propose that design and inference about design are complementary processes, grounded in a capacity to mentally simulate goal-directed actions. We tested this using "Overcooked"-style kitchens where participants either judged what a kitchen was designed for (Study 1) or designed kitchens for cooks with varying goals and beliefs (Study 2). In Study 1, participants inferred that kitchens were designed for tasks the layout made easier to complete, consistent with the prediction of a simulation-based computational model. In Study 2, participants made designs that helped cooks efficiently complete their task, adjusting their choices when cooks faced uncertainty about which task to perform. Together, these findings point towards a study of design as a cognitive activity grounded in the same mechanisms that support planning and social reasoning.

Repository structure

├── project_code/
│   ├── R/                          # Statistical analysis
│   ├── python/                     # Data processing and model pipelines
│   ├── bash/                       # bash and slurm scripts
│   └── experiments/                # jsPsych web experiments
├── data/                           # Behavioral and model data
├── stimuli/                        # Experimental stimuli
├── figures/                        # Generated figures
├── gym-cooking/                    # Simulation framework (submodule)
└── environment.yml                 # Conda environment

See the individual READMEs for more details:

• data/data-readme.md — description of behavioral and model data files
• stimuli/stimuli-readme.md — description of experimental stimuli
• project_code/python/python-readme.md — data preprocessing and simulation pipelines
• project_code/R/R-readme.md — statistical analyses and figure generation
• project_code/bash/bash-readme.md — bash and slurm scripts

Set up

The project uses Python 3 and R. We recommend using conda to set up the analysis environment:

conda env create -f environment.yml
conda activate design-inference

Experiments

Pre-registrations for each study can be found on the Open Science Framework:

• Inference Study — effort conditions
• Inference Study — inference conditions
• Design Study

Inference Study

In Study 1, participants viewed "Overcooked"-style kitchen layouts and judged what the kitchen was designed for (e.g., which dish, or how many cooks).

Example inference trial.

Demos for each condition:

• Cooks inference
• Dish inference
• Cooks effort
• Dish effort

Code for this experiment can be found in project_code/experiments/s1_design_inference.

Design Study

In Study 2, participants designed kitchens for cooks with varying goals and beliefs by placing furniture in the environment.

Example cooking trial (task familiarization).

Example design trial.

Demos for each condition:

• Cook knows which dish to make
• Cook does not know which dish to make

Code for this experiment can be found in project_code/experiments/overcooked_design.

Empirical analyses

All paper figures, tables, and statistics are generated by a single R Markdown file. Pre-computed data files are included so you can knit without running simulations.

Step 1: Download the brms model cache (~4.5 GB) from OSF and unzip it into data/overcooked_design/model_cache/. If you skip this step, the Rmd will attempt to refit the models from scratch, which takes several hours.

Step 2: Knit the analysis notebook:

cd project_code/R
Rscript -e "rmarkdown::render('analysis_cogsci2026.Rmd')"

For information about reproducing behavioral data preprocessing and simulation model runs, see project_code/python/python-readme.md.

For details on the statistical analyses, see project_code/R/R-readme.md.

External resources

• OSF project: https://osf.io/xqvs5/overview
• gym-cooking framework: https://github.com/cicl-stanford/gym-cooking

CRediT author statement

What is a CRediT author statement?

(left blank for peer review)