Agent Deception via Polynomial Path Planning

Overview

Deceptive path planning enables an intelligent agent to conceal its true intentions while appearing to pursue an alternative goal. Deception is a crucial tool for misleading and confusing adversaries, especially in sectors such as security, transportation, and surveillance, where the ability to conceal true intentions may lead to significant advantages. This repository contains the implementation of our deceptive polynomial path planner (DPPP). As shown in Figure 1, DPPP generates deceptive behaviors that guide an agent towards a target destination while simultaneously misleading an adversarial observer into predicting a false goal.

Figure 1: Overview of DPPP. The left side shows the deception factors used by the planner, including goal/decoy proximity (top row), goal/decoy alignment (bottom left), and the agent's apparent goal inferred from trajectory trends. As shown on the right, an adversary must consider these factors when determining the agent's intentions, while the agent strategically exploits these factors to generate deceptive plans.

Authors

• Nolan B. Gutierrez
• Brian M. Sadler
• William J. Beksi

Citation

If you find this project useful, then please consider citing our work.

@article{gutierrez2025agent,
  title={Agent deception via polynomial path planning},
  author={Gutierrez, Nolan B and Sadler, Brian M and Beksi, William J},
  journal={Engineering Applications of Artificial Intelligence},
  volume={159},
  pages={111205},
  year={2025},
  publisher={Elsevier}
}

Installation

Requirements

• Ubuntu 22
• Miniconda3

Environment Setup

After installing Miniconda3, create a new conda environment with Python 3.10:

$ conda create -n deception_env python=3.10 -y
$ conda activate deception_env

Then, clone the repository:

$ git clone https://github.com/robotic-vision-lab/Agent-Deception-Via-Polynomial-Path-Planning.git
$ cd Agent-Deception-Via-Polynomial-Path-Planning

Finally, install the required Python packages:

$ pip install -r requirements.txt

Getting Started

After installing the environment, you can test DPPP by running one of the example configurations:

$ python deceptive_polynomials/test.py --degree 5 --start_location [5.5,0] --goal [1.5,9.5] --alternative_goals "[[9.5,9.5]]" --short_on --ambiguity_on --reg_beta 100000 --curvature_on --beta 100 --reg_on --alt_angle_beta 1000

See the Usage section below for more options.

Usage

For batch execution or parameter sweeps, try the following example:

$ for i in {1..10}; do python deceptive_polynomials/test.py --degree $i --beta 0 --alternative_goals "[[9.5,9.5]]"  --circle_location "(7,-8)" --start_location [5.5,1.5] --goal [1.5,9.5] --circle_beta 100   --short_on --obs_on  --title "Degree_\$i"  --ambiguity_on --alt_angle_beta 1000  --curvature_on  --reg_beta 100000 ; done

This command runs DPPP 10 times, within a shell loop, with varied output titles. It is useful for generating multiple trajectories under different configurations. The command was used to generate the deceptive paths shown in Figure 2.

Figure 2: Visualization of deceptive paths generated by DPPP using different polynomial degrees. Each path is evaluated using the following metrics: distance traveled (DT), alignment deception index (ADI), distance deception index (DDI), alignment deception impact index (ADII), and distance deception impact index (DDII). Lower values of ADI, DDI, ADII, and DDII indicate higher deceptiveness, while lower DT values indicate greater efficiency.

You can modify the input parameters to suit your specific use case or scenario. Here's a brief explanation of the input parameters:

--degree: Degree of the polynomial to be fitted
--beta: Regularization parameter
--alternative_goals: Alternative goal locations
--circle_location: Location of the circle used in the path planning
--start_location: Start location of the robot/vehicle
--goal: True goal location
--circle_beta: Circle constraint regularization parameter
--short_on: Enable/disable the shortest path constraint
--obs_on: Enable/disable the obstacle constraint
--title: Title for the output file
--ambiguity_on: Enable/disable the ambiguity constraint
--alt_angle_beta: Strength of the alignment constraint with the decoy goal (encourages trajectories to appear goal-aligned)
--curvature_on: Enable/disable the curvature constraint
--reg_beta: Regularization parameter for the trajectory fitting
--point_beta: Point constraint regularization parameter
--traj_folder_prefix: Prefix for the trajectory folder name

After each run, the resulting trajectories are stored in the trajectories/ folder by default. To visualize all of the trajectories (e.g., Figure. 2) using matplotlib, execute the following command:

$ python deceptive_polynomials/utils/plot_trajectories.py --folder trajectories

References

This project utilizes code from the following project:

• PythonRobotics https://github.com/AtsushiSakai/PythonRobotics

Acknowledgments

This research was supported by the U.S. Department of Education's Graduate Assistance in Areas of National Need (GAANN) Fellowship. Initial development and early evaluation of DPPP was conducted during an internship at the U.S. Army Combat Capabilities Development Command Army Research Laboratory (DEVCOM ARL), as part of the DoD's HBCU/MI Summer Research Internship Program.

License