Multi-agent Reinforcement Learning

Course Description

In recent years, reinforcement learning (RL) has shown great potential in solving sequential decision-making problems, such as game playing or autonomous driving, where supervised signals can be sparse. Interestingly, many of the decision-making scenarios where RL has shown great potential involve multiple stakeholders, hereafter called agents, that can potentially act and learn independently of each other. These scenarios are studied in the subfield of RL called multi-agent reinforcement learning (MARL). Multi-agent RL is considerably more challenging than single-agent RL due to the fact that an RL agent in a multi-agent setting needs to account for the presence of other agents. In this course, we will outline some of these challenges, and showcase why multi-agent environments call for different algorithmic approaches than those designed for single-agent settings. We will cover both theoretical and practical aspects of MARL.

Course tools

  • Python
  • Tensorflow (Keras)


  • Markov Decision Processes
  • Deep Learning
  • Reinforcement Learning (basic)

Level of complexity of course



Dr. Goran Radanovic

Ph.D. in Computer Science from the Swiss Federal Institute of Technology in Lausanne (EPFL).

He is a research group leader at the Max Planck Institute for Software Systems (MPI-SWS), where he leads the multi-agent systems group. He is generally interested in studying AI systems, and more specifically in the design and analysis of systems with intelligent and self-interested agents. Prior to joining MPI-SWS, he was a postdoctoral researcher at Harvard University.

Fields of interests: Multi-agent Systems, Incentive Mechanism Design, Reinforcement Learning