Reactive synthesis of action planners
An increase in the level of autonomy marks one of the fundamental focuses of current robotic systems. This involves the ability of a robot to reason about its environment and plan its motion in order to carry out assigned tasks. For all tasks, it generally involves abstractions into discrete, logical actions, where each discrete action defines a particular capability of the robot.
The problem of synthesis of correct-by-construction action planners has been considered in this work. Action Description Language (ADL) is used to model the actions. These ADL definitions are then translated to Linear Temporal Logic (LTL). LTL based specifications are further used for the reactive synthesis of the strategy.
This work largely focuses on expressiveness which consists of a definition of the actions and system/environment behavior. Classical ADL semantics cannot handle multiple agents or non-determinism. A natural extension of ADL (referred to as ADLnE in this document) has been proposed which can handle dynamic environments, non-determinism, and multiple agents.
The proposed work can be seen as an extension to generic search based action planners. One such A* search-based method, Goal Oriented Action Planner (GOAP) has been considered which is based on ADL semantics and is limited by deterministic, single agent modeling. Through examples, it has been established that for deterministic, single agent and static (or at best quasi-static) systems, the proposed strategy matches that of GOAP. For dynamic and multi-agent situations, a reactive action plan is synthesized (if feasible) that is guaranteed to satisfy the formal specification, i.e. achieve the goal.