Off-campus UMass Amherst users: To download dissertations, please use the following link to log into our proxy server with your UMass Amherst user name and password.
Non-UMass Amherst users, please click the view more button below to purchase a copy of this dissertation from Proquest.
(Some titles may also be available free of charge in our Open Access Dissertation Collection, so please check there first.)
Achieving flexibility for autonomous agents in dynamic environments
Abstract
Recent Artificial Intelligence research in robot planning has indicated the potential importance of what has been called situated action in enabling autonomous agents to achieve and maintain their goals in dynamic, real-time environments. This approach involves defining a set of reactions that at all times dictate the behavior of the agent in relation to the current state of the world, without making reference to explicit, internal models. In this dissertation we demonstrate one architecture (called $\rm\sc{PLASTYC}$) for integrating purely reactive agency with a more classical, model-based planning component. Our aim is to show that ways in which such a combined architecture can gain from the benefits of both approaches, while not being limited by the shortcomings of either. The $\rm\sc{PLASTYC}$ architecture incorporates a model-based strategic planning component (the reflective module) on top of a pre-existing set of reactions (the reactive module). The reactive module alone is able to produce a moderate level of performance. The reflective module uses the blackboard model of inference in which a skeletal planning regimen is implemented. External actions are taken as a function of the dual influences from the reactive and reflective modules. If there is no action dictated by the reflective module then the only influence is that of the reactive module. The reflective module itself controls the relative influence of reactive and reflective action intentions. The reflective module incorporates other features intended to enhance its abilities to perform in real-time domains. One of these is a mechanism by which multiple internal goals of the agent can be pursued in parallel, but with the computational resources of the agent allocated according to the rated importance of the goals being pursued. We constructed a rich and complex simulated environment in which to examine these ideas empirically. A large number of experiments were conducted in which various aspects of the $\rm\sc{PLASTYC}$ agent and/or its environment were systematically varied. Among other things, these experiments indicate that the need for incorporating a reflective component is increased by the degree of "discontinuity" in the environment relative to an agent's reactions.
Subject Area
Computer science|Artificial intelligence
Recommended Citation
Day, David Spencer, "Achieving flexibility for autonomous agents in dynamic environments" (1991). Doctoral Dissertations Available from Proquest. AAI9132838.
https://scholarworks.umass.edu/dissertations/AAI9132838