Adapt the work from Keller et al. (PROST) and Sanner (RiDDLeSim) to a deep reinforcement learning tool developed at Saarland University (in process).
For this purpose, I implement a python-scipt to merge the starting of RiDDLeSim and Prost in one script executing all together.
The communication is implemented via "named pipes" according to the scheme of the learner.
In general you need to have the following:
- git in order to clone this repository
To run prost, you need to install or fulfill the following requirements:
- Linux is you operating system
- flex and bison are installed to parse the input
- cmake and c++ compiler (recommended is gcc)
- the BDD library BuDDy
You can install most of the dependencies with the following command
sudo apt install git g++ cmake bison flex libbdd-dev
Prost interacts with a server called RiDDLeSim that simulates an environment based on the specification of a .rddl-file. To be able to run this Server you need need to install java:
sudo apt install default-jre (install Java runtime environment)
To install BuDDy, download the tar.gz file from the link above and extract it. Then you need to execute the following calls (in the folder containing the unpacked buddy files):
./configure(optionally set--prefix=path/to/installation_targetin case that you don't want to install it to the default location (in case you're not admin))make && make install
If BuDDy is not found automatically, you can additionally set an environment variable
$BDD_ROOT which points to your BuDDY installation (folder containing buddy lib and include folders).
To install RiDDLeSim you need to do the following (without any guarantee for success):
cd /path/to/rddl/rddlsim(switch to directory within this repo)./compile(compile rddlsim)
Because of the execution of this repo, it is required that you set an environment
variable $RDDLSIM_ROOT that points to /path/to/rddl/rddlsim, e.g. by adding
export RDDLSIM_ROOT=/path/to/rddl/rddlsim
to the file ~/.bashrc. It has to point to the base directoy of the rddlsim subdirectory.
To finally install PROST, you need to do the following:
cd /path/to/rddl/prost(switch to directory within this repo)./build.py(compile the planner)
Both, RiDDLeSim and PROST, can be executed on its own as given in the PROST-README.md or by the combined script which is recommended as it is easier for this task.
As we use PROST only as the parser and interacting environment for a learner and
not as a planner, you only have to give the instance to the Parser. So, the arguments
are mainly the same as for RiDDLeSim and can be seen in the help of the script run.py.
The main arguments you have to provide are :
- -b: the folder containing the environment and its problem instances (located in prost/testbed/benchmarks)
- -i: the concrete instance to solve (locaed in the folder given above)
An exemplary prompt to execute the program would be
./run.py -b ./prost/testbed/benchmarks/elevator-2011/ -i ./prost/testbed/benchmarls/elevator-2011/elevators_inst_mdp__1.rddl
Roman Joeres
based his work thankfully on the work of the PROST core developers
and
- Augusto Blaas Correa
- Dorde Relic
- Florent Teichteil-Koenigsbuch
- Jendrik Seipp
- Joris Scharpff
- Max Grüner
- Patrick Eyerich
- Scott Sanner
and for the RiDDLeSim-Server
- Scott Sanner