Ensemble Programming for Multipotent Systems

Abstract

The orchestration and controlling of groups of robots, i.e., programming ensembles, is a complicated task to do. Deciding and defining what the ensemble needs to do in which order often requires in-depth problem domain specific knowledge. Further, an ensemble programmer needs to put much effort into the design, implementation, and evaluation of distributed algorithms to coordinate multi-robot executions. Most commonly, expertise for the problem domain and programming knowledge is not united in the same person. To tackle this, we propose Maple, an approach for a Multi-Agent Programming Language for Ensembles. Using Maple simplifies an ensemble programmer's typical work-flow by providing generalized solutions for solving technical tasks in ensemble programming. Maple is a graphical ensemble programming language based on the formalism of hierarchical task networks that enables the online generation of ensemble programs. Maple also enables non-technical domain experts to generate ensemble programs for specific use-cases.

Maple-Swarm: Programming Collective Behavior for Ensembles by Extending HTN-Planning

“…Achieving a general approach becomes even more complex as tasks show a high versatility or the user requires the ensemble to act in different problem domains. With our approach Maple [12], we already developed a task orchestration and execution framework for multipotent robot ensembles [15] having the potential to fill that gap. In multipotent systems, robot ensembles being homogeneous at design time can become heterogeneous concerning their capabilities at run-time by combining physical reconfiguration on the hardware level with self-awareness [10].…”

“…On task layer, we evaluate the user-specified problem domain definition against the current state of the world the system is currently aware of and generate plans for this situation with an automated planner. We integrated this automated planner in our approach for a multi-agent script programming language for multipotent ensembles (Maple) [12]. There, we extend the approach of hierarchical task network (HTN) planning [5] for defining a problem domain and generating plans.…”

“…We further need to enable the task-allocation mechanism to deal with agent groups. In Maple [12], we can transform generated plans into tasks that address specific agents at any time (e.g., directly after planning). Now, we need to perform this transformation with respect to the current situation at run-time before we allocate them to actual robots.…”

“…For creating a HT N and the partial plans ρ part it contains, the designer can use all elements of our extended HTN planning approach from [12], i.e., compound nodes (cn), primitive nodes (pn), world state modification nodes (ws), re-planning nodes (rp), as well as our concept for looped execution of nodes. A partial plan ρ part thus consists of nodes containing information the designer requires the system to execute, e.g., ρ part 1 := [pn 1 , pn 2 , ws, rp] ∈ HT N .…”

Modeling and Execution of Coordinated Missions in Reconfigurable Robot Ensembles

“…The deduction of abstract tasks using self organizing algorithms for agent missions is covered in Kosak et al [9] and combined capabilities in Eymueller et al [10]. While our previous work [11] focuses on the coordination of whole ensembles, the BDL covers the definition of robot programs, using the robots' sets of capabilities in a generic way. By using generic specifications that can be instantiated at runtime, we need no details on specific hardware implementations at design time.…”