FeAR

Feasible Action-Space Reduction (FeAR) for Causal Responsibility in Multi-Agent Spatial Interactions

Github repository for the publication : https://ebooks.iospress.nl/volumearticle/64281.

The paper proposes a metric (FeAR) for causal responsibility in multi-agent spatial interactions. The FeAR metric was implemented in a grid world simulation that was created to model multi-agent spatial interactions.

---

---

Contents

Main Files

File	Function
GWorld.py	Grid World Simulation Environment
Responsibility.py	Calculate FeAR Values
FeAR_Sims.ipynb	Run Simulations with GWorld and calculate FeAR values
AnalysisOf_FeAR_Sims.py	Visualise and Analyse the results from FeAR_Sims

Helper Files

File	Function
Agent.py	Agents used in GWorld
PlotGWorld.py	Plotters for GWorld
Scenario4FeARSims.json	JSON file with Scenarios to be used in FeAR_Sims
ScenarioBuilder.ipynb	To generate new scenarios for FeAR_Sims
Instances2plot.json	JSON file containing the list of instances for which plots will be generated by AnalysisOf_FeAR_Sims.py

Instructions to view results

• Run AnalysisOf_FeAR_Sims.py to see the UI that plots the results
• Use ↑ and ↓ arrow keys om the keyboard to scroll through different instances
• Use ← and → arrow keys om the keyboard to change the groups of instances shown
• Use (PAGEDOWN or RIGHTSHIFT) and (PAGEUP or LEFTSHIFT) to change records.
• Pressing P adds the current instance to Instances2plot.json
• Pressing G generates plots for the instances in Instances2plot.json and saves them to \Plots\
• Press S to toggle between 'Sequential View' and 'Extremes View'
  • In 'Extremes View' instances are sorted in order of the values of FeAR (Starts here by default).
  • In 'Sequential View' instances are sorted by temporal order.
    • To see how situations evolve over time, and
    • To inspect every single instance irrespective of the FeAR values.

Instructions to run simulations

• To run simulations and to calculate FeAR values use the notebook FeAR_Sims.ipynb.
• Detailed instructions are included in the notebook.
• FeAR_Sims.ipynb uses scenarios defined in Scenario4FeARSims.json.
• To generate new scenarios for FeAR_Sims use the notebook ScenarioBuilder.ipynb

Supplementary Files and Data

• https://osf.io/5s9m2/

Authors

• Ashwin George - PhD student @ TU Delft

Funding and support

• TU Delft AI Initiative

Reference

A. George, L. Cavalcante Siebert, D. Abbink, and A. Zgonnikov, “Feasible Action-Space Reduction as a Metric of Causal Responsibility in Multi-Agent Spatial Interactions,” in Frontiers in Artificial Intelligence and Applications, K. Gal, A. Nowé, G. J. Nalepa, R. Fairstein, and R. Rădulescu, Eds., IOS Press, 2023. doi: 10.3233/FAIA230349. https://ebooks.iospress.nl/volumearticle/64281.

Copyright notice

Technische Universiteit Delft hereby disclaims all copyright interest in the program “FeAR” (Feasible Action-Space Reduction) written by Ashwin George. Fred van Keulen, Dean of Mechanical, Maritime and Materials Engineering

© 2023, Ashwin George, HERALD Lab, TU Delft