Second Order State Hallucinations (SOSH) for Multi-Agent Systems

Laksh Patel (Illinois Mathematics and Science Academy), Akhilesh Raj (Vanderbilt University)

Overview

Second Order State Hallucinations (SOSH) is a novel methodolgy for mitigating attacks in formation control of multi-agent systems. Traditional mulit-agent systems, upon error, experience cascading faults throughout the system. SOSH, utilizing residual analysis, allows each agent to detect faults in the system within a threshold. Then, the network topology is updated to exclude the attacked node(s). Now, as the system lacks the attacked node(s), SOSH comes into action, approximating the attacked node(s) positions with both velocity and acceleration. The depth of approximation (second order) allows for practical use in search-and-rescue, platooning, traffic control, and military applications.

Above is an example of SOSH preventing cascading errors on the unaffected nodes (node 1, 3, 4).

Talks & Awards

• NCSSS 2025 Student Research Conference
  Awarded a fully funded trip to present SOSH at the National Consortium of Secondary STEM Schools.
• 24th Annual High School Research Symposium
  Presented SOSH and received the People’s Choice Award.
• 63rd Illinois Junior Science and Humanities Symposium
  Presented SOSH research.
• 3rd International Mathematics and Statistics Student Research Symposium
  Invited to deliver a talk on SOSH methodology.

🚀 Getting Started

Prerequisites

• Python 3.8+

  pip install -r code/requirements.txt
  
  

• MATLAB (for running analysis/analysis_extensive.m)
• C++17 (with Eigen & matplotlib-cpp for compiling analysis/analysis_full.cpp)

Clone & Navigate

git clone https://github.com/yourusername/sosh-project.git
cd sosh-project

Generate Dataset

python analysis/create_csv.py --config analysis/experiment_config.yaml

This will produce:

• results/all_positions.csv
• results/aggregated_metrics.csv

Analysis

• MATLAB Open and run analysis/analysis_extensive.m to generate figures under results/figures/.

• C++

  g++ -std=c++17 analysis/analysis_full.cpp -I/path/to/eigen -lpython3.x -o analysis_full
  ./analysis_full

• Jupyter Notebook (Optional) Launch analysis/analysis_notebook.ipynb for interactive exploration.

Animation & Visualization

python code/main.py

Runs the SOSH simulation animation with robust detection and second‐order hallucination.

📋 Requirements

# code/requirements.txt
numpy>=1.21.0
matplotlib>=3.4.0
pandas>=1.3.0

📄 License

Released under the MIT License. See LICENSE for details.