Dynamic domain reduction for multi-agent planning

摘要

We consider a scenario where a swarm of arbitrary unmanned vehicles (UxVs) are used to satisfy a multitude of diverse, spatially distributed objectives. The UxVs strive to determine an efficient schedule of tasks to service the objectives while operating as a swarm. We focus on developing autonomous high-level planning, where low-level controls are leveraged from previous work in distributed motion, target tracking, localization, and communication algorithms. We take a Markov decision processes (MDP) approach to develop a multi-agent framework that can extend to multi-objective optimization and human-interaction for swarm robotics. Utilizing state and action abstractions, we introduce a hierarchical algorithm, Dynamic domain reduction for multi-agent planning, to enable multi-agent planning for large multi-objective environments. Simulated results show significant improvement over using a standard Monte Carlo tree search in an environment with large state and action spaces.