New website coming soon
Please note that over the next month our website will be changing. Go to cecs.anu.edu.au/newsite for more information.
An Agent Architecture for Structured Uncertain Environments
Zahra Zamani (ANU)CS HDR MONITORING AI Group
TIME: 11:30:00 - 12:00:00
LOCATION: RSISE Seminar Room, A105 with Pizza
My talk is aimed at investigating a solution to building cognitive agents able to operate in a realistic environment. The agents can use the models programmed by the designer and at the same time learn to behave optimally according to the environment closely resembling the real world. The agent's knowledge about its surroundings is uncertain and although receiving input from the environment can help the agent obtain a better understanding of its current situation, the source of input can itself have some degree of noise. We consider this uncertainty in our probabilistic models designed before run-time and modified by parameters during the agent framework runs. The agent surroundings can be considered as a structured environment and the knowledge will be in the form of structures of terms and relations among them. The features of the state space will be represented in a rich higher-order Modal language. Most software systems concentrate on well known architectures such as the Belief- Desire- Intension (BDI) model, but here a different approach well known in the Robotics context, but not explored in depth for cognitive agent application, is used. We define an architecture based on motion and sensor models defined for tracking and localizing a robot in the real world. These models are refined by learning parameters to show environment characteristics. Once the models are ready the agent can use them to select actions to achieve its goal. To track the state space, the Bayes filter will be applied to a mixture of discrete and continuous distributions. Examples from the Trading Agent Competition will be provided to show this framework. Technical results on Hidden Markov Models, Linear dynamic systems and parameter estimation within this game is provided.