Housekeeping with Multiple Autonomous Robots: Representation, Reasoning, and Execution
Domestic robots capable of doing basic household tasks are becoming more ubiquitous than ever with the recent advancements in cognitive robotics. One of the interesting tasks for such robots is tidying up the house autonomously by relocating the objects in the environment. This task incorporates several challenges. For instance, robots should know that a book is generally located in a bookshelf, i.e. robots should have some commonsense knowledge, and be able to reason with it. Another challenge is generating high-level symbolic plans which respect the geometric constraints of the environment so that they can be performed by actual robots. In addition, robots should be able to collaborate with each other to overcome some complex tasks in a multi-robot system.
In this study, we initially represent the planning domain in action language C+, and use the causal reasoner CCalc to compute plans. Then, we reformulate our domain in terms of Answer Set Programming (ASP), and make use of several ASP solvers with different capabilities, such as iClingo and dlvhex, to reason on this domain with the planning purpose. On top of these planners, we implement our novel execution monitoring algorithm in order to recover the robots from possible plan failures, and show its applicability via simulation. We provide experimental results regarding the performance of the different reasoners we use for planning, as well as the plan failure frequencies of different approaches with respect to the representation of geometric constraints.
Erdi Aker is master's student in Computer Science with an emphasis in Artificial Intelligence and Cognitive Robotics. He received his BS from Sabanci University. His research interest is knowledge representation and reasoning for housekeeping robots.