CMRADAR

Overview
CMRADAR is a complete agent with capabilities ranging across the full spectrum of calendar management, from natural language processing of incoming scheduling-related emails, to making autonomous scheduling decisions, to negotiating with other users, to user interfacing and visualization. Although many research issues remain, we believe CMRADAR is the first end-to-end agent for automated calendar management.
 

Technical Details

A key contribution of the design of CMRadar is the specification of a basic representation, called a Template, for communicating calendar scheduling related information. The Template data structure is used as the language for the communication between the components in CMRadar and as the "glue" that binds them together. In addition, Templates are also used to normalize unformatted natural language emails into a machine readable format. We offer the Template data structure as a flexible approach to the general design of a meeting scheduling agent.

The CMRadar architecture contributes a modular design in which the core scheduling functions of the agent are separated from the multiagent aspects of calendar management. Rather than an approach that tightly couples schedule optimization and negotiation, CMRadar has a separate Manager component which handles the sending and receiving of messages from other agents and more generally, manages the negotiation with others. The Manager then communicates via Templates with a separate Scheduler component that handles the core optimization problems. We found that this modular architecture facilitates the integration of existing scheduling systems and indeed, a core component of CMRadar is the Ozone scheduler originally designed for and used in several real-world logistics planning domains.

The primary underlying emphasis of the Radar project is to learn to improve performance, adapt to unexpected situations and to customize to different users. The emphasis on learning is reflected in our design of the CMRadar architecture in which all components read and write data to a central knowledge base that can be used by a separate learning process to provide feedback to the decision making components. Indeed, it is the need to collect real-world data to support learning that drives our development of a complete end-to-end agent.