You are here

Prof. Kraus' Lab

Prof. Kraus' Lab

Head - Computer Science Lab



Tel: 972-3-531-8762

Computer Science: Multi-Agent Systems

Prof. Sarit Kraus of the Department of Computer Science and the Leslie and Susan Gonda (Goldschmied) Multidisciplinary Brain Research Center is best known for her pioneering work in imparting intelligence upon computerized agents.  Kraus introduced the aspect of individualism to the field of multiple agent systems by developing protocols and strategies for individual agents that need to collaborate with each other.

“Shared Plans” – a TeamWork Model for Collaborative Planning

Together with Prof. Barbara Grosz of Harvard, Kraus developed a reference theory for collaborative planning (a TeamWork model) called “SharedPlans,” which provides specification for the design of collaboration-capable agents and a framework for identifying and investigating fundamental questions about collaboration. 

It specifies the minimal conditions for a group of agents to have a joint goal, the group and individual decision-making procedures that are required, the way the agents' mental states and plans can evolve over time, and various other important relationships among the agents, e.g., teammates, subcontractors, etc.

Given the extensiveness of SharedPlans and its rigorous specifications, it has been the basis for many other works and has been widely adopted in other fields (e.g. robotics or human-machine interaction). 

Automated Negotiation Systems

Kraus founded the field of automated systems that professionally negotiate with people, and for many years was virtually the only researcher working in this important and challenging area. Her work in automated negotiation between computerized agents led to the development of a breakthrough agent that negotiated both in the United States and in Lebanon.  These agents combine machine learning techniques for general opponent modeling with formal decision making approaches. They also deploy the concept of social utility functions that consist of factors both from the agent point of view and from the human partner’s point of view. All of these agents were evaluated empirically in extensive studies involving hundreds of human subjects.

Automated Security Systems

In conjunction with colleagues at the University of Southern California, Kraus has developed a system called “ARMOR” - Assisted Randomized Motoring Over Routes – which is based on the randomization of countermeasures – those activities through which security forces attempt to foil the efforts of would-be terrorists.  ARMOR casts these problems as Bayesian Stackelberg games, allowing the agent to appropriately weigh the different actions in randomization as well as uncertainty over adversary types. ARMOR has been successfully deployed since August 2007 at the Los Angeles International Airport (LAX) to randomize checkpoints on the roadways entering the airport and canine patrol routes within the airport terminals.

Adversarial Patrolling

Her work on adversarial patrolling with Dr. Noa Agmon and Prof. Gal Kaminka considered the problem of a multi-robot patrol and an adversary’s attempt to penetrate the patrol path without being detected. While others have focused on the frequency of the robots' visits at each point on the patrol path, Kraus and her collaborators have emphasized the need to take the adversary into consideration. For situations where there are insufficient robots and no guarantee that the adversary will be deterred, they proposed a model for increasing the probability of success by applying a non-deterministicpatrol framework.

Last updated on 29/5/14