This paper introduces an approach that allows an agent to exploit inherent patterns of interaction in its environment, so-called dynamics, to achieve its objectives. The approach extends the standard treatment of planning and (re) action in which part of the input to the plan generation algorithm is a set of basic actions and perhaps some domain axioms. Real world actions are typically difficult to categorize consistently and are highly context dependent. The approach presented here takes as input a procedural model of the agent's environment and produces as output a set of action descriptions that capture how the agent can exploit the dynamics in the environment. An agent constructed with this approach can utilize context sensitive actions, "servo" style actions, and other intuitively efficient ways to manipulate its environment. A process-algebra based representation, RS, is introduced to model the environment and the agent's reactions. The paper demonstrates how to analyze an RS environment model so as to automatically generate a set of potentially useful dynamics and convert these to action descriptions. The output action descriptions are designed to be input to an Interval Temporal Logic based planner. A series of examples of reaction construction drawn from the kitting robot domain is worked through, and the prototype implementation of the approach described. © 1995.
Lyons, D. M., & Hendriks, A. J. (1995). Exploiting patterns of interaction to achieve reactive behavior. Artificial Intelligence, 73(1–2), 117–148. https://doi.org/10.1016/0004-3702(94)00058-9