HR: 17:15h
AN: IN14A-05 [Abstracts]
TI: Planning and Scheduling for Environmental Sensor Networks
AU: * Frank, J D
EM: frank@email.arc.nasa.gov
AF: NASA Ames Research Center, Mail Stop N269-3, Moffett Field, CA 94035-1000
United States
AB:
Environmental Sensor Networks are a new way of monitoring the environment. They
comprise autonomous sensor nodes in the environment that record real-time data,
which is retrieved, analyzed, integrated with other data sets (e.g. satellite
images, GIS, process models) and ultimately lead to scientific discoveries.
Sensor networks must operate within time and resource constraints. Sensors have
limited onboard memory, energy, computational power, communications windows and
communications bandwidth. The value of data will depend on when, where and how
it was collected, how detailed the data is, how long it takes to integrate the
data, and how important the data was to the original scientific question.
Planning and scheduling of sensor networks is necessary for effective, safe
operations in the face of these constraints. For example, power bus limitations
may preclude sensors from simultaneously collecting data and communicating without
damaging the sensor; planners and schedulers can ensure these operations are
ordered so that they do not happen simultaneously.
Planning and scheduling can also ensure best use
of the sensor network to maximize the value of collected science data. For
example, if data is best recorded using a particular camera angle but it is
costly in time and energy to achieve this, planners and schedulers can
search for times when time and energy are available to achieve the optimal camera angle.
Planning and scheduling can handle uncertainty in the problem specification;
planners can be re-run when new information is made available, or can generate
plans that include contingencies. For example, if bad weather may prevent the collection
of data, a contingent plan can check lighting conditions and turn off
data collection to save resources if lighting is not ideal.
Both mobile and immobile sensors
can benefit from planning and scheduling. For example, data collection on otherwise
passive sensors can be
halted to preserve limited power and memory resources and to reduce the costs of
communication.
Planning and scheduling is generally a heavy consumer of time, memory and energy
resources. This means careful thought must be given to how much planning and
scheduling should be done on the sensors themselves, and how much to do
elsewhere. The difficulty of planning and scheduling is exacerbated when
reasoning about uncertainty. More time, memory and energy is needed to solve
such problems, leading either to more expensive sensors, or suboptimal plans.
For example, scientifically interesting events may happen at
random times, making it difficult to ensure that sufficient resources are
availanble. Since uncertainty is usually lowest in proximity to the sensors
themselves, this argues for planning and scheduling onboard the sensors.
However, cost minimization dictates sensors be kept as simple as possible,
reducing the amount of planning and scheduling they can do themselves. Furthermore,
coordinating each sensor's independent plans can be difficult.
In the full presentation, we will critically review the planning and
scheduling systems used by previously fielded sensor networks.
We do so primarily from the perspective of the computational sciences, with a focus
on taming computational complexity when operating sensor networks.
The case studies are derived from sensor networks
based on UAVs, satellites, and planetary rovers. Planning and scheduling considerations
include multi-sensor coordination,
optimizing science value, onboard power management, onboard memory,
planning movement actions to acquire data, and managing communications.These case studies offer lessons for
future designs of environmental sensor networks.
DE: 9805 Instruments useful in three or more fields
DE: 9810 New fields (not classifiable under other headings)
DE: 9815 Notices and announcements
DE: 9820 Techniques applicable in three or more fields
SC: Earth and Space Science Informatics [IN]
MN: Fall Meeting 2005