For mobile robots to perform complex missions, it may be necessary for them to plan with incomplete information and reason about the indirect effects of their actions. Answer Set Programming (ASP) provides an elegant way of formalizing domains which involve indirect effects of an action and recursively defined fluents. In this paper, we present an approach that uses ASP for robotic task planning, and demonstrate how ASP can be used to generate plans that acquire missing information necessary to achieve the goal. Action costs are also incorporated with planning to produce optimal plans, and we show how these costs can be estimated from experience making planning adaptive. We evaluate our approach using a realistic simulation of an indoor environment where a robot learns to complete its objective in the shortest time.

PDF, PS, HTML

@inproceedings{AAAISSS14-yang, title={Planning in Answer Set Programming while Learning Action Costs for Mobile Robots}, author={Fangkai Yang and Piyush Khandelwal and Matteo Leonetti and Peter Stone}, month={March}, url="http://www.cs.utexas.edu/users/ai-lab?yang:aaaisymp14", year={2014} }