Making Friends in the Dark: Ad Hoc Teamwork Under Partial Observability
This addresses the challenge of collaborative AI in uncertain environments, but it is incremental as it builds on prior ad hoc teamwork work with new assumptions.
The paper tackles the problem of ad hoc teamwork under partial observability, where an agent must assist unknown teammates without access to their actions or a reward signal, and shows that their model-based approach is effective and robust across 70 POMDPs from 11 domains.
This paper introduces a formal definition of the setting of ad hoc teamwork under partial observability and proposes a first-principled model-based approach which relies only on prior knowledge and partial observations of the environment in order to perform ad hoc teamwork. We make three distinct assumptions that set it apart previous works, namely: i) the state of the environment is always partially observable, ii) the actions of the teammates are always unavailable to the ad hoc agent and iii) the ad hoc agent has no access to a reward signal which could be used to learn the task from scratch. Our results in 70 POMDPs from 11 domains show that our approach is not only effective in assisting unknown teammates in solving unknown tasks but is also robust in scaling to more challenging problems.