There have been many projects in AI planning. The approach taken usually depends on the scope of the desired solution. For real-world applications, computing resources and time are often very limited, so the solution must be highly tailored to the application.
NASA's Remote Agent, Autonomous Sciencecraft Experiment, and the Mars rovers Spirit and Opportunity are examples of application-specific automated planning. These missions are particularly significant because they represent a considerable level of trust in the technologies that are deployed, since millions of taxpayer dollars were at stake.
Projects with a more general approach include the General Problem Solver developed by Newell and Simon in 1957 and its successor, Soar. Soar is currently being developed at the University of Michigan, and claims DARPA among its supporters. Soar has evolved from a planning system to an approach to general AI with planning at its core.
Patrick Doyle's web page on planning provides a long list of specific projects from the last half of the 20th century, and the approaches used by their developers. The Association for the Advancement of Artificial Intelligence (AAAI) web site also contains many web links, both current and historical, on the page covering planning.
Automated Planning
planning (verb)
1. To formulate a scheme or program for the accomplishment, enactment, or attainment of: plan a campaign.
2. To have as a specific aim or purpose; intend: They plan to buy a house.
- American Heritage Dictionary
AI (or automated) planning is a set of goal-seeking heuristics whereby the current state of a system is analyzed and a future desired state, along with a sequence of intermediate states, is selected. This discipline has uses that span the entire range of computing – from robotic sampling systems for space and sea, to industry and gaming. It draws heavily on cognitive science, as researchers attempt to create better planning systems based on the apparent methods employed by humans and animals every day. If we consider robotics as the discipline that reduces the human workload and danger level in many of life's activities, then automated planning is and will increasingly become the mechanism that manages the robotic labor force.
Approaches to automated planning can be thought of as lying on a continuum ranging from zero to complete automation. At one end are systems that are completely reactive in nature - when some input changes, the system's control law causes one or more outputs to be altered to achieve a new desired state. The solutions tend to be highly mathematical, and analysis involves a lot of graphs. The planning for these systems is done by engineers and operators.
Often, however, there are conditions that cannot be modeled mathematically, and a few hard rules need to be introduced to prevent errant behavior. Progression along that track leads to a more AI-like approach, whose ultimate goal is the creation of systems that are able to react to fundamental changes in their environment or their information about it (e.g. fault tolerance beyond the classical notion of stability).
Current automated planning applications seem to be focused on reducing the need for human involvement in the daily activities of robotic systems. For example, the concept of autonomy in modern spacecraft has expanded so that mission controllers might only have to specify a set of coordinates to have a satellite image a particular site. NASA's Automated Sciencecraft Experiment seeks to expand the concept even further by allowing the satellite to essentially choose its own target by analyzing past images to determine the best coordinates for future images.
Subscribe to:
Post Comments (Atom)
Bibliography
- Bernard, D., Dorais, G., Fry, C., Gamble, E., Kanefsky, B., Kurien, J., Millar, W., Muscettola, N., Nayak, P., Pell, B., Rajan, K., Rouquette, N., Smith, B., and Williams, B. “Design of the Remote Agent Experiment for Spacecraft Autonomy.” Proceedings of the IEEE Conference on Aerospace. Snomass, Co (1998). April 15, 2008. http://ti.arc.nasa.gov/projects/remote-agent/publications.php
- Nau, Dana. “Current trends in automated planning.” AI Magazine 28.4 (2007): 43–58.
- Paley, Derek. Personal Interview. 25 Apr. 2008.
- Pistore, Marco and Varde, Moshe. “The Planning Spectrum – One, Two, Three, Infinity.” Journal of Artificial Intelligence Research. 30 (2007): 101-132
No comments:
Post a Comment