The Fifth Annual AAAI Mobile Robot Competition and Exhibition was held in Portland, Oregon, in conjunction with the Thirteenth National Conference on Artificial Intelligence. The competition consisted of two events: (1) Office Navigation and (2) Clean Up the Tennis Court. The first event stressed navigation and planning. The second event stressed vision sensing and manipulation. In addition to the competition, there was a mobile robot exhibition in which teams demonstrated robot behaviors that did not fit into the competition tasks. The competition and exhibition were unqualified successes, with nearly 20 teams competing. The robot competition raised the standard for autonomous mobile robotics, demonstrating the intelligent integration of perception, deliberation, and action.

Net-ID (San Francisco, Calif.) has been involving the deployment of the and performance management as well awarded a $500,000 grant by the National Army's vast resources. Intelligent Optimization (St. Louis, software for the rapid analysis and Mo.) has developed optimizor, a neural StockSmart (Dallas, Tex.) has integrated classification of the large number of network program designed to help hospital quest server, an intelligent searchand-discovery DNA and protein sequences produced staff members make more efficient The AIbased creates an optimal schedule that its web site. Visitors to StockSmart's software will be used to mine reflects an institution's own preferences web site can use Resumix (Sunnyvale, Calif.) has won a among thousands of mutual funds. The contract calls for Technology Office. Users will have installed in each regional location of systems in a real-time, interactive environment.

The Third International Conference on Artificial Intelligence Planning Systems (AIPS-96) was held in Edinburgh, Scotland, from 29 to 31 May 1996. The main gathering of researchers in AI and planning and scheduling, the conference promoted the practical applications of planning technologies. Details of the conference papers and sessions are provided as well as information on the Defense Advanced Research Projects Agency -- Rome Laboratory Planning Initiative.

The National Aeronautics and Space Administration (NASA) is being challenged to perform more frequent and intensive space-exploration missions at greatly reduced cost. Nowhere is this challenge more acute than among robotic planetary exploration missions that the Jet Propulsion Laboratory (JPL) conducts for NASA. This article describes recent and ongoing work on spacecraft autonomy and ground systems that builds on a legacy of existing success at JPL applying AI techniques to challenging computational problems in planning and scheduling, real-time monitoring and control, scientific data analysis, and design automation.

This article was reprinted with permission from "ACM Computing Surveys: 28(4): December 1996, 653-670. original is available from ACM.

This article is the content of an invited talk given by the authors at the Thirteenth National Conference on Artificial Intelligence (AAAI-96). The piece begins with a short history of the competition, then discusses the technical challenges and the political and cultural issues associated with bringing it off every year. We also cover the science and engineering involved with the robot tasks and the educational and commercial aspects of the competition. We finish with a discussion of the community formed by the organizers, participants, and the conference attendees. The original talk made liberal use of video clips and slide photographs; so, we have expanded the text and added photographs to make up for the lack of such media.

Elucidation of the relationship between TD learning and dynamic programming (DP) has provided good theoretical insights (Barto et al., 1995). However, conventional TD algorithms were based on discrete-time, discrete-state formulations. In applying these algorithms to control problems, time, space and action had to be appropriately discretized using a priori knowledge or by trial and error. Furthermore, when a TD algorithm is used for neurobiological modeling, discrete-time operation is often very unnatural. There have been several attempts to extend TD-like algorithms to continuous cases. Bradtke et al. (1994) showed convergence results for DPbased algorithms for a discrete-time, continuous-state linear system with a quadratic cost. Bradtke and Duff (1995) derived TD-like algorithms for continuous-time, discrete-state systems (semi-Markov decision problems). Baird (1993) proposed the "advantage updating" algorithm by modifying Q-Iearning so that it works with arbitrary small time steps.

We describe the reinforcement learning problem, motivate algorithms whichseek an approximation to the Q function, and present new convergence results for two such algorithms. 1 INTRODUCTION AND BACKGROUND Imagine an agent acting in some environment. At time t, the environment is in some state Xt chosen from a finite set of states. The agent perceives Xt, and is allowed to choose an action at from some finite set of actions. Meanwhile, the agent experiences a real-valued cost Ct, chosen from a distribution which also depends only on Xt and at and which has finite mean and variance. Such an environment is called a Markov decision process, or MDP.

A method for incorporating context-dependent phone classes in a connectionist-HMM hybrid speech recognition system is introduced. Amodular approach is adopted, where single-layer networks discriminate between different context classes given the phone class and the acoustic data. The context networks are combined with a context-independent (CI) network to generate context-dependent (CD) phone probability estimates. Experiments show an average reduction in word error rate of 16% and 13% from the CI system on ARPA 5,000 word and SQALE 20,000 word tasks respectively. Due to improved modelling, the decoding speed of the CD system is more than twice as fast as the CI system.

In Tesauro's 1992 landmark work on TD--gammon, he showed that the temporal dif--