This paper addresses the problem of how a group of agents cooperating on a complex plan with interdependent actions can coordinate their scheduling and execution of those actions, particularly in domains where actions may fail or have uncertain durations.  If actions fail (or fail to meet their deadlines), the repercussions for the rest of the team's plan can be dramatic.  This paper presents a pro-active strategy, called Just-in-Time Backfilling (JIT-BF), that agents can use to increase the fault tolerance of their interdependent schedules by identifying actions in danger of failing and inserting redundant (or back-up) actions into their schedules.  The insertion of redundant actions can be done locally (i.e., by the agent whose action is in danger of failing) or through negotiations with the rest of the team. The computations performed by agents following the JIT-BF strategy depend on probabilistic models of action durations and the ``quality'' achieved by successfully executing actions.  The paper presents an experimental evaluation of the JIT-BF strategy within a simulated real-time dynamic environment that demonstrates that teams using the pro-active JIT-BF strategy significantly out-perform teams that rely solely on reactive strategies.

In this paper an application of the metaheuristic Ant Colony Optimization to optimal planning is presented. It is well known that finding out optimal solutions to planning problem is a very hard computational problem. Approximate methods do not guarantee either optimality or completeness, but it has been proved that in many applications they are able to find very good solutions, often close to optimal ones. Since one of the most performing stochastic method for combinatorial optimization is ACO, we have decided to use this technique to design an algorithm which optimizes plan length in propositional planning. This algorithm has been implemented and some empirical evaluations have been performed. The results obtained are encouraging and show the feasibility of this approach.

In our experiments with four well-known systems for solving partially observable planning problems  (Contingent-FF, MBP, PKS, and POND), we were greatly surprised to find that they could only solve problems with a small number of contingencies. Apparently they were repeatedly trying to solve many combinations of contingencies at once, thus unnecessarily using up huge amounts of time and space. This difficulty can be alleviated if the planner can maintain focus on the contingency that it is currently trying to solve. We provide a way to accomplish this by incorporating focusing information directly into the planning domain's operators, without any need to modify the planning algorithm itself. This enables the above planners to solve larger problems and to solve them much more quickly. We also provide a new planner, FOCUS, in which focusing information can be provided as a separate input. This provides even better performance by allowing the planner to utilize more extensive focusing information.

Planning has belonged to fundamental areas of AI since its beginning and sessions on planning are an integral part of major AI conferences. By generating activities necessary to achieve some goal, planning is also closely related to scheduling that deals with allocation of activities to scarce resources. Although the planning and scheduling communities are somehow separated, both areas have interacted more and more in recent years, especially when dealing with real-life problems. This FLAIRS special track attempts to make the conference attractive for the planning community, a traditional part of the AI family, and also the scheduling community -- especially for those using AImotivated solving techniques such as constraint satisfaction. FLAIRS 2008 hosted the first special track on AI planning and scheduling.

Computer Science has a bad reputation among non-CS majors. This paper describes three assignments from a gentle introduction to programming course for non-majors that uses robots and simple game programming as a hook to get students interested in the subject. In each of the assignments presented, what might be considered a trivial twist to an instructor was a key factor in making an otherwise standard project into something that is more engaging.

The field of artificial intelligence needs to attract new researchers to the field to continue current explorations and look for novel approaches to tomorrow's problems. One approach involves providing students with learning tools that excite their imagination and help them obtain an appreciation for what artificial intelligence can do. The tools described here are used in an undergraduate course at Sam Houston State University. They include heuristic-driven search in a potential game's terrain map, reinforcement learning in a tank battle game, and game tree search techniques in tic-tac-toe.

In this paper, we examine Robot Defense, a computer game that serves as a pedagogical platform for students to explore methods typically covered in an Introductory Artificial Intelligence course. Robot Defense is the synergistic outcome of two NSF funded Course, Curriculum, and Laboratory Improvement (CCLI) projects and was first presented in (Wallace, Russell and Markov 2008). The primary contribution of this paper is to discuss the implementation of the Robot Defense platform and the outcome of its first use in the classroom.

We present a little crawling robot with a two DOF arm that learns to move forward within about 15 seconds in real time. Due to its small size and weight the robot is ideally suited for classroom demonstrations as well as for talks to the public. Students who want to practice their knowledge about reinforcement learning and value iteration can use a wireless connection to a PC and monitor the internal state of the robot such as the value function or the reward table. Due to its adaptivity, depending on the surface properties of the underground the robot may surprise its audience with unexpected but efficient walking policies. The GUI is open source and the robot hardware is available as a kit from the authors.

LEGO Mindstorms robots are very popular with colleges and universities for teaching computer concepts and programming.  These robots elicit excitement in students and provide a nontrivial, real-world platform for exploring algorithmic concepts.  We created a simple algorithmic language, called Robolang, and wrote a translator that turns it into Lejos code, a variant of Java that can be run on the RCX version of the LEGO Mindstorms robots.  Seeing that students were eager to explore programming with the RCX robots at home, we wrote a graphical simulator to visualize actions of our penbot, a configuration of the RCX robot that we used in most assignments.  Using an emulator approach, we intercept the ROM calls to the RCX's hardware made by the TinyVM, the stripped-down Java Virtual Machine that runs compiled Java bytecodes.  Our system then forwards these calls to a software model that represents the actual robot hardware.  The software model creates the graphics to mimic the penbot using Java2D. This approach greatly simplified coding by capitalizing on existing software, namely the Java compiler and the JVM.  Students can program either in Robolang or in actual Lejos and use the simulator to visualize the actions of the robot acting as a sort of visual debugger.

Using robots as part of any curriculum requires careful management of the significant complexity that physical embodiment introduces. Students need to be made aware of this complexity without being overwhelmed by it, and navigating students through this complexity is the biggest challenge faced by an instructor.  Achieving this requires a framework that allows complexity to be introduced in stages, as students' abilities improve. Such a framework should also be flexible enough to provide a range of application environments that can grow with student  sophistication, and be able to quickly change between applications.  It should be portable and maintainable, and require a minimum of overhead to manage in a classroom. Finally, the framework should provide repeatability and control for evaluating the students' work, as well as for performing research. In this paper, we discuss the advantages of a mixed reality approach to applying robotics to education in order to accomplish these challenges.  We introduce a framework for managing mixed reality in the classroom, and discuss our experiences with using this framework for teaching robotics and AI.