Agent Adaptability

To operate effectively in open environments agent systems must be adaptive. The RETSINA agents have been designed explicitly to be adaptive. Adaptation is behavior of an agent in response to unexpected events or dynamic environments. Examples of unexpected events include the unscheduled failure of an agent, an agent's computational platform, or underlying information sources. Examples of dynamic environments include the occurrence of events that are expected but it is not known when (e.g., an information agent may reasonably expect to become at some point overloaded with information requests), events whose importance fluctuates widely (e.g., price information on a stock is much more important while a transaction is in progress, and even more so if certain types of news become available), the appearance of new information sources and agents, and finally underlying environmental uncertainty (e.g., not knowing beforehand precisely how long it will take to answer a particular query). In RETSINA, adaptivity is exhibited at multiple levels, from the application level of interacting agents down to selecting different individual method executions based on changing conditions. Thus, in RETSINA, the following types and levels of adaptivity have been implemented :

  • Adaptivity of an individual agent. An individual agent can exhibit adaptivity relative to a variety of internal reasoning processes:
    • agent communication
    • agent coordination
    • agent planning
    • agent scheduling
    • agent execution monitoring
  • Adaptivity regarding learning from interactions with a user or another agent..
  • Adaptivity at the organizational level of the multi-agent system. Adaptivity with respect to multi-agent system performance.
Software adaptivity at all these levels results in:
  1. agent systems that support runtime selection of agents for collaborative performance of a task, thus reconfiguring themselves to improve system performance and robustness
  2. agent systems that have a model of their own specifications and behavior and can alter their behavior based on changes in the environment
  3. agent systems that introspect, diagnosing anomalies at runtime and repairing them


Multidisciplinary University Research Initiative (MURI)
Principal Investigator: Katia Sycara
Sponsored by: Office of Naval Research (ONR)
ONR Contact: Michael Shneier
© 1998 Carnegie Mellon Robotics Institute

Carnegie Mellon School of Computer Science Carnegie Mellon Robotics Institute Office of Naval Research Carnegie Mellon School of Computer Science Carnegie Mellon Robotics Institute Office of Naval Research Carnegie Mellon MURI Carnegie Mellon University Carnegie Mellon School of Computer Science Carnegie Mellon Robotics Institute Office of Naval Research Carnegie Mellon MURI Introduction MURI Agents in Teams Reusable Agents Information Filtering Middle Agents Interagent Operability Agent Autonomy Agent Coordination Agent Adaptability Conflict Negotiation MODSAF Aircraft Maintenance Warren Carnegie Mellon School of Computer Science Carnegie Mellon Robotics Institute DefenseLINK Office of Naval Research MURI Welcome University of Pittsburgh School of Information Sciences