|
I am interested in the integration of planning and learning methods to solve decision-making problems, in domains with uncertainty and/or partial observability. This means the agent (or robot) must be able to quickly assess some abstract value of some given action prospects, for the current state, despite having only limited knowledge and time, and choose an action that is likely to produce the desired results. While there are methods that solve tasks like these with good numerical performance (optimality), they are simply impractical for real-time planning in robots and other similar agents due to their high computational cost.
For this reason we are looking for solutions that yield satisficing behavior, and that are able to handle task-planning and problem-solving in dynamic, variable domains by quickly focusing on a few relevant alternatives.
|
