This package contains end-to-end tools for the verification and synthesis of data driven systems using Gaussian process regression and IMDP abstraction.
Must be installed using Julia 1.6+. Installation will fail for versions 1.5 and below. Build tools must be installed for certain packages to work. This limitation will be addressed soon. In the meantime, you can install essential build tools in Ubuntu via apt-get install build-essential.
From the Julia REPL,
- Press
]to activate the Pkg prompt - Add the package in development mode,
dev https://github.com/aria-systems-group/CautiousEngine.jl.git - Activate the package environment,
activate ~/.julia/dev/CautiousEngine - Instantiate the environment,
instantiate
Check if everything worked by pressing backspace to go back to the julia> prompt, and try using CautiousEngine. If this succeeds without error, try running an example:
include("~/.julia/dev/CautiousEngine/examples/verification-example-single.jl")
Since this package is not public yet, simply clone and add the repository to the Julia load path (or add it to startup.jl) via:
push!(LOAD_PATH, "path/to/CautiousEngine.jl").
If ~/.julia/config/startup.jl does not exist, you can create it. A good location for packages under development is ~/.julia/dev.
Next, use the following commands to activate the project environment and then instantiate the project and install dependencies:
pkg"activate path/to/CautiousEngine.jl" then pkg"instantiate".
This will install all the required packages.
This package depends on the bmdp-tool here: https://github.com/aria-systems-group/bmdp-tool
The tool should be compiled using Make and the synthesis executable moved to a location on the user executable path e.g. /usr/local/bin. CautiousEngine does not use the included MATLAB tools.
Full documentation is in progress, but here are some notes to get started.
Full examples are provided in examples/ for synthesis and verification applications. The synthesis example initially uses a small number of points to test that the package is working. Copy this file and modify it as needed.
Note how measurement and process noise are handled in the scripts. The synthesis procedure can only use process noise, and the verification procedure can only use measurement noise.
The synthesis example uses an optional known component of the dynamics. This is currently limited to knowing x_k in x_{k+1} = x_k + g(x_k) but not g(x_k).
Contributions are welcome, especially those related to testing, code optimization, documentation, etc.