Mission Specification Compliant Recovery for Robotic Autonomous Vehicles

Code for the paper "SpecGuard: Specification Aware Recovery for Robotic Autonomous Vehicles from Physical Attacks", CCS'2024.

Key Features

• Mission specification compliant safe recovery for robotic autonomous vehicles (RAV).
• Integrate Deep-RL with autopilot software such as PX4 and ArduPilot.
• Training pipeline to learn a recovery control policy that maintains mission specification compliance even under physical attacks.
• Multi-agent training to optimize the recovery control policy.

The structure is as follows:

drone
|
└─── configs    <--- configuration files
└─── missions   <--- mission details
└─── modified_px4   <--- PX4 hooks to connect recovery policy with autopilot


rover
|
└─── configs    <--- configuration files
└─── missions   <--- mission details
└─── modified_ardupilot   <--- Ardupilot hooks to connect recovery policy with autopilot

environment <--- AirSim environment binaries

docker
|
└─── modified_px4   <--- PX4 hooks to connect recovery policy with autopilot
└─── modified_ardupilot   <--- Ardupilot hooks to connect recovery policy with autopilot
└─── DroneDockerfile   
└─── RoverDockerfile
...  

requirements.txt

Dependencies and Prerequisites

We developed and tested on Ubuntu 20.04 and Python 3.8. The following are the dependencies and prerequisites.

1. AirSim: AirSim is built on Unreal engine, and it provides physically and visually realistic simulations of the environment. Follow the instructions provided by Microsoft to build and setup the AirSim simulator.
2. Create a Virtual Environment:
   • Create a new Python virtual environment: python3.8 -m venv .env.
   • Activate the virtual environment: source .env/bin/activate.
   • Install the required dependencies: pip install -r requirements.txt.
3. Install Environments: The environments directory contains scripts for downloading RAV operating environments (e.g., suburban, urban city block, downtown, etc.).
4. Setup: Docker or Local Installation: Additional setup steps for different RAV types are in drone and rover directory. We also provide docker containers to run our solutions, details are in docker/ directory.

Citation

If you find our work useful in your research, please consider citing:

@inproceedings{dash2024recovery,
      title={SpecGuard: Specification Aware Recovery for Robotic Autonomous Vehicles from Physical Attacks}, 
      author={Dash, Pritam and Chan, Ethan and Pattabiraman, Karthik},
      booktitle = {ACM Conference on Computer and Communications Security (CCS)},
      year={2024}
}