We propose that a planner should be provided with an explicit model of its own planning mechanism, and show that linking a planner's expectations about the performance of its plans to such a model, by means of explicit justification structures, enables the planner to determine which aspects of its planning are responsible for observed performance failures.
Planning in real-time offers several benefits over the more typical techniques of implementing Non-Player Character (NPC) behavior with scripts or finite state machines. NPCs that plan their actions dynamically are better equipped to handle unexpected situations. The modular nature of the goals and actions that make up the plan facilitates reuse, sharing, and maintenance of behavioral building blocks. These benefits, however, come at the cost of CPU cycles. In order to simultaneously plan for several NPCs in real-time, while continuing to share the processor with the physics, animation, and rendering systems, careful consideration must taken with the supporting architecture. The architecture must support distributed processing and caching of costly calculations. These considerations have impacts that stretch beyond the architecture of the planner, and affect the agent architecture as a whole. This paper describes lessons learned while implementing real-time planning for NPCs for F.E.A.R., a AAA first person shooter shipping for PC in 2005.
In this paper, we investigate the hypothesis that plan recognition can significantly improve the performance of a case-based reinforcement learner in an adversarial action selection task. Our environment is a simplification of an American football game. The performance task is to control the behavior of a quarterback in a pass play, where the goal is to maximize yardage gained. Plan recognition focuses on predicting the play of the defensive team. We modeled plan recognition as an unsupervised learning task, and conducted a lesion study. We found that plan recognition was accurate, and that it significantly improved performance. More generally, our studies show that plan recognition reduced the dimensionality of the state space, which allowed learning to be conducted more effectively. We describe the algorithms, explain the reasons for performance improvement, and also describe a further empirical comparison that highlights the utility of plan recognition for this task.
Eric B. Baum 1 NEC Research Institute, 4 Independence Way, Princeton NJ 08540 eric@research.NJ.NEC.COM Abstract The point of game tree search is to insulate oneself from errors in the evaluation function. The standard approach is to grow a full width tree as deep as time allows, and then value the tree as if the leaf evaluations were exact. This has been effective in many games because of the computational efficiency of the alpha-beta algorithm. A Bayesian would suggest instead to train a model of one's uncertainty. This model adds extra information in addition to the standard evaluation function. Within such a formal model, there is an optimal tree growth procedure and an optimal method of valueing the tree. We describe how to optimally value the tree, and how to approximate on line the optimal tree to search.
Leonardo Garrido Ram6n Brena Centro de Inteligencia Artificial, Tecnol6gico de Monterrey Abstract This paper presents recent results of our experimental work in quantifying exactly how useful is building models about other agents using no more than the observation of others' behavior. The testbed we used in our experiments is an abstraction of the meeting scheduling problem, called the Meeting Scheduling Game, which has competitive as well as cooperative features. The agents are selfish, and use a rational decision theoretic approach based on the probabilistic models that the agent is learning. We view agent modeling as an iterative and gradual process, where every new piece of information about a particular agent is analyzed in such a way that the model of the agent is further refined. We propose a framework for measuring the performance of different modelling strategies and establish quantified lower and upper limits for the performance of any modeling strategy. Finally, we contrast the performances of a modeler from an individual and from a collective point of view, comparing the benefits for the modeler itself as well as for the group as a whole. Introduction Katia Sycara The Robotics Institute, Carnegie Mellon University Several approaches in the field of multiagent systems (MAS) (Durfee 1991; Wooldridge & Jennings 1995) make heavy use of beliefs as an internal model of the world (Bratman 1987) One form of belief of particular importance in multiagent systems are the agent's beliefs about other agents (Vidal & Durfee 1997b). This kind of belief could come from preexisting knowledge base (a kind of"prejudice"), or could be inferred from observing others' behavior. The purpuse of a modelling activity could be to benefit a specific agent, in the case of "selfish" agents, or to improve the performance of a group as a whole, in the case of cooperative agents -or even a combination of both.