CMSC 477/677 - Spring 2005

Discussion Questions for Class #6, February 15

Reading: Apex website; Remington et al., "Using Apex/CPM-GOMS to Develop Human-Like Software Agents, AAMAS 2003; "Paper Covers Rock: An Apex Tutorial."

Questions:

What was the primary motivation behind developing the Apex system? How does this motivation differ from those of the BDI, Soar, and ACT-R architectures?
What is CPM-GOMS? What is the difference (if any) between CPM-GOMS, hierarchical task network (HTN) planning, and the universal subgoaling of Soar?
What are the main characteristics of the task environments towards which Apex is targeted? To what extent are these characteristics similar to, or different from, the task characteristics targeted by BDI architectures and/or Soar?
Under what circumstances are the "operator independence assumptions" mentioned by Remington et al. violated? Can you come up with some example domains where the independence assumptions might not hold? How do you think the other architectures we've studied (BDI, Soar, ACT-R) handle such domains?
Do you think Roshambo is an interesting multitasking environment?
The Apex tutorial describes several strategies for decision-making in Roshambo, ranging from the trivial ("always play rock") to the nearly trivial ("always try to beat the other player's last move"). Can you propose some non-trivial strategies? Maybe we'll talk more about Roshambo later in the semester...