FireCommander: An Interactive, Probabilistic Multi-agent Environment for\n Heterogeneous Robot Teams

摘要

The purpose of this tutorial is to help individuals use the\n\\underline{FireCommander} game environment for research applications. The\nFireCommander is an interactive, probabilistic joint perception-action\nreconnaissance environment in which a composite team of agents (e.g., robots)\ncooperate to fight dynamic, propagating firespots (e.g., targets). In\nFireCommander game, a team of agents must be tasked to optimally deal with a\nwildfire situation in an environment with propagating fire areas and some\nfacilities such as houses, hospitals, power stations, etc. The team of agents\ncan accomplish their mission by first sensing (e.g., estimating fire states),\ncommunicating the sensed fire-information among each other and then taking\naction to put the firespots out based on the sensed information (e.g., dropping\nwater on estimated fire locations). The FireCommander environment can be useful\nfor research topics spanning a wide range of applications from Reinforcement\nLearning (RL) and Learning from Demonstration (LfD), to Coordination,\nPsychology, Human-Robot Interaction (HRI) and Teaming. There are four important\nfacets of the FireCommander environment that overall, create a non-trivial\ngame: (1) Complex Objectives: Multi-objective Stochastic Environment,\n(2)Probabilistic Environment: Agents' actions result in probabilistic\nperformance, (3) Hidden Targets: Partially Observable Environment and, (4)\nUni-task Robots: Perception-only and Action-only agents. The FireCommander\nenvironment is first-of-its-kind in terms of including Perception-only and\nAction-only agents for coordination. It is a general multi-purpose game that\ncan be useful in a variety of combinatorial optimization problems and\nstochastic games, such as applications of Reinforcement Learning (RL), Learning\nfrom Demonstration (LfD) and Inverse RL (iRL).\n