Internship: Applying reinforcement learning to calibrate a multi-agent serious game — with a case study on waste management

The ACROSS International Laboratory in Hanoi (http://across-lab.org) is dedicated to the use of computer models and simulations for the participatory management of large socio-ecosystems. Within this framework, ACROSS is actively involved in the development of the GAMA modeling platform (http://gama-platform.org), a powerful tool for agent-based simulation (Taillandier et al., 2019). 

In recent years, a growing trend in agent-based simulation has been the development of serious games, where human players interact with simulation environments to learn and experiment with complex systems. A recent example is the RAC game, developed with the GAMA platform, in which groups of players take on the role of village leaders making decisions about waste management (Biré et al., 2025). The game models four villages that share both local and common objectives.

Designing such a game requires a substantial calibration phase to define appropriate parameter values — such as action payoffs, success thresholds, and coordination mechanisms — so that the game remains challenging yet educational, effectively conveying the intended message (e.g., that coordination among villages is crucial for success). In this context, reinforcement learning (RL) offers a promising approach to assist with game calibration. By exploring the policy space for different parameter configurations, RL can help identify “optimal” strategies, which can then be analyzed to assess whether they align with the intended pedagogical objectives of the game.

A first connection between GAMA and popular RL libraries such as Gymnasium and PettingZoo has already been established. The objective of this internship is to implement and evaluate the GAMA–RL integration on a real-world use case, namely for the calibration of the RAC game. This work may involve further extending and refining the existing framework to ensure robustness and adaptability to complex multi-agent interactions.

The internship will be conducted in collaboration with INRAE Toulouse, where researchers specialized in reinforcement learning will provide their expertise. The project strongly emphasizes transdisciplinarity, fostering collaboration among experts in agent-based simulation, artificial intelligence, and sustainability science.

The internship is structured around two main objectives:

1. Extending the Python Library for GAMA-RL Coupling

• Review and build upon existing work on the Python library, that is GAMMA-RL, that connects GAMA with Gymnasium and PettingZoo interfaces
• Improve its usability and performance for multi-agent reinforcement learning (MARL).
• Ensure compatibility with various GAMA simulation models.

2. Application for RAC

• Adapt the existing model within GAMA to ensure it is fully compatible and flexible for multi-agent reinforcement learning (MARL) experiments.
• Develop the Python modules that enable the exploration of the policy space across different parameter configurations, allowing the four virtual agents (representing the villages) to learn and achieve the best possible collective and individual outcomes. Particular attention will be given to the design and selection of appropriate MARL methods for the RAC case study — for instance, defining how agents’ learning algorithms balance cooperation and competition, how reward functions reflect both local and global objectives, and how coordination mechanisms can emerge through learning
• Analyse the policies learned in relation to the pedagogical objective of the game and compare them with the human players’ policies.

• Good programming skills in Python.
• Knowledge of multi-agent systems and reinforcement learning.
• Interest in sustainability-related research topics.
• Familiarity with GAMA is a plus, but not mandatory.
• Ability to work in a collaborative and interdisciplinary research environment.

• Location: ACROSS Laboratory, Hanoi, Vietnam
• Duration: 4 to 6 months, starting in April
• Supervision: The internship will be supervised entirely at ACROSS.

This internship offers a unique opportunity to work at the intersection of AI, simulation, and sustainability science, contributing to cutting-edge research in multi-agent reinforcement learning.

Biré, L., Phung, Q. N., Taillandier, P., Phung, D. A., Nguyen, N. D., & Drogoul, A. (2025). RÁC: A Serious Agent-Based Simulation Game to Drive Discussion on Waste Management in Vietnamese Irrigation Systems. Journal of Artificial Societies and Social Simulation, 28(2).

Taillandier, P., Gaudou, B., Grignard, A., Huynh, Q. N., Marilleau, N., Caillou, P., … & Drogoul, A. (2019). Building, composing and experimenting complex spatial models with the GAMA platform. GeoInformatica, 23, 299-322.

If you are interested and would like to apply, please send a CV and a cover letter to across@tlu.edu.vn. All applications will be processed on a rolling basis (no specific deadline) and interviews will be held with the most interesting profiles.