Automated planners are increasingly being integrated into online acting systems. The integration may, for example, embed a domain-independent temporal planner in a manufacturing system (e.g., the Xerox printer application) or autonomous vehicles (e.g., a planetary rover or an underwater glider). The integration may resemble something more like an "acting and planning stack" where an automated planner produces an activity or task plan that is further refined by an actor before being executed by the execution platform of the actor, such as, a reactive controller (e.g., robotics). Or, the integration may be a domain-specific policy that maps states to actions (e.g., reinforcement learning). Models for planning and execution can be same or different; the planning model can define context-dependent actions schema for online (re-)planning or can just specify flexibility to be handled separately at execution time. Online learning may or may not be involved, and may include adjusting or augmenting the model, determining when to repair versus replan, learning to switch policies, etc. A specific focus of these integrations involves online deliberation and managing the execution of actions, bringing to the foreground concerns over how much computational effort planning should invest over time.
In any of these systems, a planner generates action sequences that are eventually dispatched to an executive, yet taking action in a dynamic world rarely proceeds according to plan. When planning assumptions are challenged during execution, or some dynamic events occur, it raises a number of interesting questions about how the system should respond and which is the scope of online deliberation versus execution. Is the "acting" side of the system responsible for a response or the "planning" side? Or do the two need to cooperate and how much? When should the activity planner abandon or preempt the current goals? Should the task planner repair a plan or replan from scratch? Should the executive adjust its current policy, switch to a new one, or learn a new policy from more relevant experience?
The fifth edition of the workshop on Integrated Planning, Acting, and Execution (IntEx) aims: (1) to provide a forum for discussing the challenges of integrating online planning, acting, and execution, and (2) to assess the potential for holding an integrated execution competitions at ICAPS. Invited topics include:
We plan a one-day workshop with a focus on discussion. In contrast to the invited speaker form, we plan to host “opposing views” debates, where two members of the community will be encouraged to debate a topic relevant to the workshop in as creative a way as possible. Small-group breakouts after the debates will address questions raised by the debate so all participants gain from the experience.
Submissions may be regular papers (up to 8 pages plus references) or short position/challenge papers (up to 4 pages plus references). All papers should conform to the AAAI formatting guidelines and style (https://www.aaai.org/Publications/Templates/AuthorKit21.zip). The papers must be submitted in a PDF format via EasyChair system (https://easychair.org/conferences/?conf=intex2021) Submissions will be reviewed by at least two referees.
We welcome existing publications from other venues that are appropriate for discussion at this workshop. Please note in the title area if this work is already accepted at another venue. If the work is under review at another venue (e.g., IJCAI-2021) please notify the organizers so we can avoid potential reviewing conflicts.
At least one author of each accepted paper must attend the workshop in order to present the paper. Authors must register for the ICAPS main conference in order to attend the workshop. There will be no separate workshop-only registration.
Submission link: https://easychair.org/conferences/?conf=intex2021