USC-CALO Project

People

 Tapana Gupta

 Emma Bowring
 Jonathan P. Pearce
 Pradeep Varakantham
 Prof. Milind Tambe

 

Past Contributors

 Prof. Rajiv T. Maheswaran

Research Summary

We are working on a project known as CALO (Cognitive Agent that Learns and Organizes), a personal assistant agent that will observe and learn from its user and the environment, assist its user with tasks, and communicate with other CALOs. This work is supported by the Defense Advanced Research Projects Agency (DARPA), through the Department of the Interior, NBC, Acquisition Services Division, under Contract No. NBCHD030010.

Our work as part of the CALO project is to explore these topics in the domain of distributed task allocation, in which personal agents must work together, either to schedule multiple events which their users must attend, or to assign tasks to users as part of a project plan. While cooperating, the agents are also attempting to honor their users' preferences and privacy concerns. and operate effectively given a dynamic, uncertain environment (new conflicts may arise, events may need to be rescheduled, new tasks may be added to a project plan, communication between agents may be unreliable, etc.)

Currently, our work falls into four threads: Distributed Constraint Optimization, Adjustable Autonomy, Privacy, and Teamwork.

Distributed Constraint Optimization

Our research in this area is concerned with constraint-based techniques for coordination between autonomous agents in dynamic, uncertain environments. In many domains, such as scheduling, multiple spacecraft coordination, disaster rescue and reconnaissance, agents must try to agree on how to meet group-level goals in a timely fashion.

Our approach is to model these types of problems as DCOPs (Distributed Constraint Optimization Problems). We are exploring modifications to existing DCOP algorithms, as well as new DCOP algorithms, and examining their performance in different types of DCOPs in different domains.

Visit our DCOP web page for code, datasets, and more information regarding DCOP.

Key papers:

Background work:

Adjustable Autonomy

Adjustable autonomy (AA) encompasses the strategies by which an agent selects the appropriate entity (itself, a human user, or another agent) to make a decision at key moments when an action is required. Our initial work on AA for CALO has focused on the use of Partially Observed Markov Decision Processes (POMDPs) to account for the uncertainties in the domain. The key point of this work is it exploits the notion of progress present in Personal Assistant Domains for enhancing the performance of these POMDPs.

Key papers:                                                                                          

Background work:

Privacy

Although members of a multi-agent team are working together toward a common goal, it is often desirable to preserve the privacy of team members. We are exploring DCOP as a testbed for evaluating privacy in collaborative multi-agent systems.

Key papers:

Teamwork

The Machinetta proxy will be used to allow multiple CALO agents to coordinate as a team. Please go here for more information about Machinetta.

Key papers:

  • N. Schurr, S. Okamoto, R. T. Maheswaran, P. Scerri and M. Tambe, "From STEAM to Machinetta: The evolution of a BDI teamwork model," In Cognition and Multi-Agent Interaction: From Cognitive Modeling to Social Simulation. Ron Sun (editor), Cambridge University Press, 2004.

Links

CALO Project:

SRI's CALO page
CALO home page.

 (account needed)

Agents research at USC:

TEAMCORE group home page
agents@usc