/FVMCTS_experiments

Experiments for https://arxiv.org/abs/2101.04788

Primary LanguageJuliaMIT LicenseMIT

FVMCTS_experiments

Setup

Ensure you have julia installed. Then clone this repository somewhere:

git clone https://github.com/rejuvyesh/FVMCTS_experiments

Ensure that you are in the experiment directory:

cd FVMCTS_experiments

Then run in the terminal:

julia --project 'using Pkg; Pkg.instantiate(); Pkg.resolve()'

Now you are all set to run the scripts for running the experiments.

SysAdmin

For evaluating the random planner on the ring sysadmin domain with global rewards for 8 agents over 100 runs, and storing the results in /tmp/logdir:

julia --project run_sysadmin.jl ring_global 8 random 100 /tmp/logdir

Similarly, for evaluating maxplus planner, with tree search depth as 3, number of iterations of tree search as 20, exploration constant as 5, and maximum number of message passing iterations as 10:

julia --project run_sysadmin.jl ring_global 8 maxplus 100 /tmp/logdir 3 20 5 10

For evaluating varel planner with tree search depth as 3, number of iterations of tree search as 20, and exploration constant as 5:

julia --project run_sysadmin.jl ring_global 8 varel 100 /tmp/logdir 3 20 5

UAV Delivery

For evaluating the random planner on problem 1 (8 agents) over 100 runs, and stroing the results in /tmp/logdir:

julia --project run_uav.jl 1 random 100 /tmp/logdir

Similarly, for evaluating the maxplus planner with tree search hyperparameters as used in the paper:

julia --project run_uav.jl 1 maxplus 100 /tmp/logdir

and varel planner:

julia --project run_uav.jl 1 varel 100 /tmp/logdir

Note: varel memory requirements are quite high and may run out of memory on larger problems.