Presenter<\/strong>: Domen \u0160OBERL Title<\/strong>: Planning with qualitative models in robotic domains<\/p>\n Abstract<\/strong>: <\/body>
\nDomen \u0160oberl is doing his PhD at the Faculty of Computer and Information Science in Ljubljana. Currently he is a teaching assistant on various computer science courses at UP FAMNIT and member of the Department of Information Sciences and Technologies.<\/p>\n
\nHumans reason qualitatively, which is one of the reasons why humans are better at implementing abstract ideas in the actual world than robots. The concept of learning human-intuitive qualitative models has largely been explored in the field of autonomous agent learning by exploration, where the learned qualitative models serve as comprehensible explanations of the agent’s findings. Presenting a qualitative model to a machine with an intention for the machine to accomplish a task, raises challenging problems for which the traditional planning methods are not well-suited.
\nIn our research we address the problem of using qualitative models in the form of Qualitative Differential Equations (QDE) for planning motion in continuous domains. Using qualitative reasoning and simulation, the robot devises a general, non-numerical plan on how to solve the problem, and executes it directly, without the need for additional numerical training or exploration. Numerical adaptation takes place during the first execution, and best performance is typically achieved from the second execution on. We demonstrate our method on some interesting robotic problems.<\/p>\n
<\/a>\n\t\t\t<\/div><\/figure>