Skip to content

This repository contains the source code for the paper titled "Uncovering Systemic and Environment Errors in Autonomous Systems Using Differential Testing"

License

Notifications You must be signed in to change notification settings

ANSWER-OSU/AIProbe

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

59 Commits
 
 
 
 
 
 
 
 

Repository files navigation

AIProbe: Uncovering Systemic and Environment Errors in Autonomous Systems Using Differential Testing

Full paper: "Uncovering Systemic and Environment Errors in Autonomous Systems Using Differential Testing", Rahil P. Mehta*, Yashwanthi Anand*, Manish Motwani and Sandhya Saisubramanian.

AIProbe is a black-box testing framework for identifying and attributing execution anomalies in autonomous agents. It systematically generates diverse environment-task configurations and uses a search-based oracle planner to determine whether the anomaly stems from the agent's model or the environment itself.

AIProbe is designed to:

  • Detect execution anomalies
  • Attribute failures to agent errors or environment errors
  • Work across discrete/continuous and single/multi-agent domains
  • Operate in black-box settings with no access to the agent’s internal model or reward function

Dependencies

Dependencies include (but are not limited to):

gymnasium
numpy
scipy
matplotlib
tqdm
torch (for PPO models)

The domains we evaluate require additional installation. To install these requirements, cd src/<DOMAIN_NAME> and follow the environment installation instructions.

Repository Contents

The source code for generating environment-task configurations and evaluating the planner and agent on these configurations are contained with src/<DOMAIN_NAME>. The domains includes ACAS Xu, Coop Navi, Bipedal Walker, Flappy Bird, and Lava.

Details of each environment and the instructions to run AIProbe on these domains are provided within the respective folders.

License

This work is released under the Creative Commons Zero v1.0 Universal (CC0-1.0) license.

About

This repository contains the source code for the paper titled "Uncovering Systemic and Environment Errors in Autonomous Systems Using Differential Testing"

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 2

  •  
  •  

Languages