Skip to content

Reinforcement learning framework to find and analyze the likeliest failures of a system under test.

License

Notifications You must be signed in to change notification settings

NASA-SW-VnV/AdaStress.jl

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

AdaStress

AdaStress is a software package that implements and extends the Adaptive Stress Testing (AST) framework, which determines the likeliest failures for a system under test.

AdaStress provides three primary services:

  • Interfaces between user simulations and the AST framework
  • Reinforcement learning-based solvers
  • Analysis and visualization tools

Prerequisites

AdaStress is written in the Julia programming language. If you do not have Julia on your computer, follow the official instructions to download the latest version. For information on using Julia, see the language documentation. To download and build the latest version of AdaStress, open the Julia REPL, and type

using Pkg
Pkg.add("AdaStress")

or type ] to enter the interactive package mode and enter the command

add AdaStress

Problem setup

To effectively make use of AdaStress, you must have a system under test (SUT) and a simulation in which the SUT interacts with a semi-stochastic environment. This means that there should be variables in your simulation that are random and behave according to a modeled probability distribution. The system should have an identifiable failure criterion and you should preferably be able to specify a measure of distance to failure, a scalar quantity which achieves its minimum at a failure event.

For instance, if you are stress-testing an aircraft collision avoidance system, your simulation might involve an encounter scenario with multiple random variables, such as pilot intent, wind, and sensor noise. The failure criterion might be the condition of two aircraft coming within a certain distance of each other. The distance to failure would then be the instantaneous distance between the aircraft. For an involved implementation of this problem, see the example in examples/cas.

AdaStress provides two basic simulation interfaces, black-box and gray-box. The type of simulation determines which solvers are may be used.

  • A black-box simulation does not reveal its environment variables and performs all updates internally. AdaStress interacts with the simulation by setting a random seed.
  • A gray-box simulation makes its environment available to the solver, which can sample the random variables directly and return the values for the simulator to use in its update step.

Your simulation must inherit from the BlackBox or GrayBox type and implement the methods found in src/interface/BlackBox.jl or src/interface/GrayBox.jl.

Further information

For more detailed instructions on using AdaStress, see the complete documentation. Example notebooks can be found in the examples directory. For background on the AST formulation, see the original paper.

License

AdaStress has been released under the NASA Open Source Agreement version 1.3, as detailed here.

Maintainers

About

Reinforcement learning framework to find and analyze the likeliest failures of a system under test.

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages