Emergence in Multi-Agent Systems: A Safety Perspective

Setup

1. Set up a virtualenv with python 3.10 or higher. You can use pyvenv or conda for this.
2. Run pip install -r requirements.txt to get requirements.
3. In case there is no study_out/ folder in the root directory, create one.

Rerunning the Experiments

The respective experiments from our paper can be reenacted in main.py. Just select the function representing the part of our experiments you want to rerun and execute it via the __main__ function.

Further Remarks

1. We use config files located in the marl_factory_grid/environment/configs and the marl_factory_grid/algorithms/rl folders to configure the environments and the RL algorithm for our experiments, respectively. You don't need to change anything to rerun the experiments, but we provided some additional comments in the configs for an overall better understanding of the functionalities.
2. Instead of collecting coins in the coin-quadrant environment our original implementation works with the premise of cleaning piles of dirt, thus it is named dirt_quadrant in the code instead. Note that this difference is only visual and does not change the underlying semantics of the environment.
3. The code for the cost contortion for preventing the emergent behavior of the TSP agents can be found in marl_factory_grid/algorithms/tsp/contortions.py.
4. The functionalities that drive the emergence prevention mechanisms for the RL agents is mainly located in the utility functions get_ordered_dirt_piles (line 91) (for solving the emergence in the coin-quadrant environment) and distribute_indices (line 165) (mechanism for two_doors), that are part of marl_factory_grid/algorithms/rl/utils.py