AAAI Publications, 2017 AAAI Spring Symposium Series

Font Size: 
Machine Learning Approach for Task Generation in Uncertain Environments
Luke Marsh, Iryna Dzieciuch, Douglas Lange

Last modified: 2017-03-20

Abstract


The command and control of unmanned vehicles is a cognitively intensive task for human operators. Efficient and successful operator performance often depends on a multitude of parameters, such as training, human abilities/factors, timing and situational awareness. Humans are required to multitask in an uncertain environment, process situational data and be able to efficiently utilize autonomous agents in multiple regions of interest. These requirements quite often result in information overload which has consequences on the success of the mission. This is currently an unsolved problem and calls for greater optimization and automation of the command and control of unmanned vehicles. The cooperative control of unmanned agents in uncertain environments has been a challenge. Many models rely on continuous-time, state-space searches in decision trees that are used for planning and execution of the mission. The methods have a number of benefits, however, this approach requires optimization of coordination of play states, by learning from environment’s temporal and spatial patterns. In addition, the rate of events in the uncertain environment is always changing, making play models inefficient under a high load of events. This paper attempts to define a space of possible models used in uncertain environments under different levels of complexity, through optimization of assignment coordination.

Keywords


Machine learning, state space optimization, complexity reduction, uncertain environments

Full Text: PDF