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

Abstract

Programming goal-oriented behavior in collective adaptive systems is complex, requires high effort, and is failure-prone. If the system's user wants to deploy it in a real-world environment, hurdles get even higher: Programs urgently require to be situation-aware. With our framework Maple, we previously presented an approach for easing the act of programming such systems on the level of particular robot capabilities. In this paper, we extend our approach for ensemble programming with the possibility to address virtual swarm capabilities encapsulating collective behavior to whole groups of agents. By using the respective concepts in an extended version of hierarchical task networks and by adapting our self-organization mechanisms for executing plans resulting thereof, we can achieve that all agents, any agent, any other set of agents, or a swarm of agents execute (swarm) capabilities. Moreover, we extend the possibilities of expressing situation awareness during planning by introducing planning variables that can get modified at design-time or run-time as needed. We illustrate the possibilities with examples each. Further, we provide a graphical front-end offering the possibility to generate mission-specific problem domain descriptions for ensembles including a light-weight simulation for validating plans.

Swarm and Collective Capabilities for Multipotent Robot Ensembles

“…In a simplified major catastrophe scenario, a firefighter might want to a) gather its ensemble of mobile robots, b) move them collectively to the area where, e.g., a gas accident happened, c) search for the source of the gas leak, and d) survey the area close to the leak (video materials on our GitHub). We can instruct our system, e.g., with our task-orchestration approach for ensembles Maple-Swarm [11]. To handle this scenario we can use the c v sw with different sets of parameters in steps a)-d) each (cf.…”