Depicting probabilistic context awareness knowledge in deliberative architectures

Abstract

Facing long-term autonomy with a cognitive architecture raises several difficulties for processing symbolic and sub-symbolic information under different levels of uncertainty, and deals with complex decision-making scenarios. For reducing environment uncertainty and simplify the decision-making process, this paper establishes a method for translating robot knowledge to a conceptual graph to later extract probabilistic context information that allows to bound of the actions present at the deliberative layer. This research develops two ROS components, one for translating robot knowledge to the conceptual graphs and one for extracting context knowledge from this graph using Bayesian networks. We evaluate these components in a real-world scenario, performing a task where a robot notifies to a user a message of an event at home. Our results show an improvement in task completion when using our approach, decreasing the planning requests by 65% and doing the task in a third of the time.