Planning and control: Thomas Dean and Michael Wellman, (Morgan Kaufmann, San Mateo, CA, 1991) 486 pages, $44.95
✍ Scribed by James Hendler
- Book ID
- 102989473
- Publisher
- Elsevier Science
- Year
- 1995
- Tongue
- English
- Weight
- 630 KB
- Volume
- 73
- Category
- Article
- ISSN
- 0004-3702
No coin nor oath required. For personal study only.
✦ Synopsis
It's a pretty frustrating thing these days to be a "Good Old Fashioned AI" researcher, as we've been branded by Searle and others. Neural network and genetic algorithms researchers are taking us to task over low-level data, our industrial partners are demanding increasingly complex real-time performance, and our funding agents are talking about high performance computing and intelligent control. Let's face it, it's a tough time for symbolic approaches-these guys demand numbers, and they want them soon! Many of us have faced this challenge using a time tested strategy-hunkering down in our research trenches and hoping that this firestorm will eventually pass. Let the engineers give them equations, logic will serve us in the future as it has in the past. Unfortunately, as AI people look to building more and more complex systems, hiding from numbers seems to be less and less successful-we must come forth and do battle, lest we are blown away by the winds of the research wars, losing the spoils to the eventual victors.
One of the places in AI where this conflict has become clear is in the area of planning and robotics. Over the past fifteen years or so, a schism has developed between the fields. AI research in planning has focused on the search-related problems of computing complete and correct solutions to conjoined-goal symbolic problems. Roboticists, on the other hand, have been focusing on problems such as the theories of kinematics and control. AI researchers have largely ignored sensors and effecters; roboticists use "planning" to mean finding paths through space, ignoring issues of long-term goals. The AI planning toolkits are filled with logics, situation calculus and the STRIPS assumption; the engineer worries about proving convergence, Kalman filtering, and adaptive control.
In the past few years, however, a small vanguard of researchers in both areas have been realizing that there is some merit to the work done by the others. A few roboticists