Humans-Robots Sliding Collaboration Control in Complex Environments with Adjustable Autonomy

摘要

Autonomous agents dealing with partial knowledge about the environment are a classical subject of study for the decision making community. Moreover, such agents sometimes have to deal with unpredictable situations, which makes any previously computed behavior useless. In this paper, we address such problems using multi-human/multi-robot interactions, where the agents evolve in a complex environment and ask humans for help when they meet unpredicted situations. We introduce a model called HHP-MDP (Human Help Provider MDP), that aims at handling the difficult situations met by the agents by using the human's help. For this purpose, we show how the agents can detect difficult situations and send different types of requests to the set of humans. The model describes how a controller can handle different requests and assign agent requests to the humans by taking into account their previously learned abilities. This controller is designed to reduce the human's cost of bother. Moreover, we show how to optimize the human's situation awareness and limit inconsistencies between her recommendations and the agent's plans.