Topics: Distributed and Multi-Agent RL
When multiple agents simultaneously learn to
cooperatively solve a problem, interactions among agents can defeat
learning. The methods in multi-agent reinforcement learning seek to
minimize such distractions and maximize cooperation towards a common
goal. Some approaches include: modeling the agents' interaction
with the environment, pre-specifying behaviors, and modeling team
goals. Significant challenges are presented by how to represent roles
within a team and the relationships between those roles.
publications