Raghav Thakar, Gaurav Dixit, Siddarth Iyer, and Kagan Tumer. 2025. Multiagent Credit Assignment for Multi-Objective Coordination. In Proceedings of the Genetic and Evolutionary Computation Conference (GECCO '25). Association for Computing Machinery, New York, NY, USA, 663–672. https://doi.org/10.1145/3712256.3726445
This code base contains all the code used in the paper, from the simulation environment to the main contribution, and the baselines as well.
Run main.py with the desired arguments to launch your choice of algorithm in your choice of environment.
Define parameters for your algorithm and environment in a .yaml inside config/.
Use ExpUtils/RoverEnvViz.py to visualise your config of the Multi-Objective Rover Domain Environment.
Example usage:
python main.py dmo rover ~/dmo_rover_data/ \
config/concentric/8ag_concentric_DMOConfig.yaml \
config/concentric/8ag_concentric_MORoverEnvConfig.yaml \
2024 dmo_rover_test 5
- nsga2: Plain NSGA-2
- dmo: Main contribution of the paper, implements credit assignment with a modified NSGA-2 algorithm (K-parent crossover)
- kpnsga2: Ablated version with no credit assignment, but still performs the K-parent crossover
- nsga2+d: Baseline from the literature
Yliniemi, L., Tumer, K. (2014). Multi-objective Multiagent Credit Assignment Through Difference Rewards in Reinforcement Learning. In: Dick, G., et al. Simulated Evolution and Learning. SEAL 2014.
If you find this repository useful, please consider citing our work:
Raghav Thakar, Gaurav Dixit, Siddarth Iyer, and Kagan Tumer. 2025. Multiagent Credit Assignment for Multi-Objective Coordination. In Proceedings of the Genetic and Evolutionary Computation Conference (GECCO '25). Association for Computing Machinery, New York, NY, USA, 663–672. https://doi.org/10.1145/3712256.3726445
Maintained by Raghav Thakar. Please raise an issue if you have an issues using this repository.