Skip to content

Modified gymnasium setup for Multi agent Reinforcement Learning w modifiable and benchmark maps

Notifications You must be signed in to change notification settings

VineetTambe/multi-agent-rl

 
 

Repository files navigation

multi-agent-rl

For overall project summary click here

Cloning Steps:

git clone --recursive -j8 [email protected]:VineetTambe/multi-agent-rl.git

Installation steps:

It is recomended that you create a python virtual environment and install the following packages.

  1. cd gym-multigrid/ && pip install -e .
  2. cd ..
  3. cd rl-baselines3-zoo/ && pip install -e . && pip install -r requirements.txt

If you face any errors while install box2d run the following -

sudo apt install swig
pip install -r requirements.txt

How to train:

cd to the rl-baselines3-zoo repository and run the following command:

python3 train.py --env multigrid-mapf-v0 -lb ./logs/ppo/ --algo ppo --env-kwargs scenario_file:\'/home/vineet/competition/Start-Kit/example_problems/warehouse.domain/warehouse_small_10.json\'
python3 train.py -tb ./logs/ppo/ --env multigrid-mapf-v0 --conf-file ./hyperparams/python/ppo_cnn_config.py --algo ppo --env-kwargs scenario_file:\'/home/admin/multi-agent-rl/Start-Kit/example_problems/warehouse.domain/warehouse_small_10.json\' agent_view_size:40

How to update training params:

The training params are located in the hyperparams folder in rl-baselines3-zoo directory - you can update the params for the algorithm being used here.

Use the following file to run the model with other python code:

cd to the rl-baselines3-zoo repository and run the following command:

You will have to modify the following lines 87 - 90 The model.predict code is at 215 - 225

python3 custom_runner.py

How to run stand alone:

cd to the rl-baselines3-zoo repository and run the following command:

python3 enjoy.py --env multigrid-mapf-v0 --algo ppo --env-kwargs scenario_file:\'/home/vineet/competition/Start-Kit/example_problems/warehouse.domain/warehouse_small_10.json\' -f logs/ --exp-id 2 --load-best

About

Modified gymnasium setup for Multi agent Reinforcement Learning w modifiable and benchmark maps

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • C++ 78.6%
  • Python 20.6%
  • Shell 0.3%
  • CMake 0.2%
  • Makefile 0.2%
  • Batchfile 0.1%