https://github.com/rejuvyesh/fvmcts_experiments
Experiments for https://arxiv.org/abs/2101.04788
https://github.com/rejuvyesh/fvmcts_experiments
multiagent-planning
Last synced: 3 months ago
JSON representation
Experiments for https://arxiv.org/abs/2101.04788
- Host: GitHub
- URL: https://github.com/rejuvyesh/fvmcts_experiments
- Owner: rejuvyesh
- License: mit
- Created: 2021-01-24T02:28:45.000Z (over 5 years ago)
- Default Branch: master
- Last Pushed: 2021-04-30T04:11:55.000Z (about 5 years ago)
- Last Synced: 2025-01-21T05:43:25.297Z (over 1 year ago)
- Topics: multiagent-planning
- Language: Julia
- Homepage: https://sites.google.com/stanford.edu/fvmcts/
- Size: 19.5 KB
- Stars: 0
- Watchers: 4
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# FVMCTS_experiments
## Setup
Ensure you have julia installed. Then clone this repository somewhere:
```bash
git clone https://github.com/rejuvyesh/FVMCTS_experiments
```
Ensure that you are in the experiment directory:
```bash
cd FVMCTS_experiments
```
Then run in the terminal:
```bash
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`:
```bash
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`:
```bash
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`:
```bash
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`:
```bash
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:
```bash
julia --project run_uav.jl 1 maxplus 100 /tmp/logdir
```
and `varel` planner:
```bash
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.