What Do You Want From Me? Adapting to the Uncertainty of Human Preferences

Pre-requisites

This code needs Python 2.7. Since it relies on Gambit 15, you need to build Gambit from source as well as its Python extension.

Module Description

• abm_analysis.py: This script generates CSV files with simulation results (simulate_and_store produces files in the format of *_experiment_results.csv) and then performs statistical analysis and writes its results to metrics.csv. Its run_parallel_simulations function is used in multiple scripts for running IMPACT+ simulations. The metrics.csv includes, per scenario:
  • Mean evacuation time.
  • Standard deviation of evacuation time.
  • Minimum evacuation time.
  • Maximum evacuation time.
  • Post-Hoc pairwise Dunn's test, against the other scenarios.
• abm_gamemodel.py: This script contains the generate_game_model function, that uses Gambit to return a game-theoretic model of the robot-survivor interaction. It is used by the abm_runner.py script.
• abm_runner.py: This is the script called within the Netlogo IMPACT+ simulation, for providing a decision for the robot interacting with a survivor. In each call to this script, we generate an instance of AutonomicManagerController equipped with a TypeAnalyser instance (see analyser.py) for decisions based on identity inference.
• abm_sensitivity.py: This scripts generates the data and plots for the sensitivity analysis of evacuation time as a function of fall length and help factor.

IMPACT+ Explainer

• IDEA robot exploring: Lime green vehicle.
• IDEA robot requesting survivor help: Magenta vehicle.
• Passengers: Black stick figures.
• First responders: Green stick figures.
• Passengers evacuating: Pink stick figures.
• Dead passengers: Grey stick figures.