The application of path finding techniques for homeland applications used in this paper was inspired by work done Li and Rus [4]. Here a versatile information system by using distributed sensor networks, i. e. hundreds of small sensors, equipped with limited memory and multiple sensing capabilities which autonomously organize and reorganize themselves as ad hoc networks in response to task requirements and to triggers from the environment.
Distributed adaptive sensor networks are reactive computing systems, well suited for tasks in extreme environments, especially when the environmental model and the task specifications are uncertain and the system has to adapt to them. A collection of active sensor networks can follow the movement of a source to be tracked, for example, a moving vehicle. It can guide the movement of an object on the ground, for example, a surveillance robot. Or it can focus attention over a specific area, for example, a fire in order to localize its source and track its spread.A sensor network consists of a collection of sensors distributed over some area that form an ad hoc network.
Each sensor is equipped with some limited memory and processing capabilities, multiple sensing modalities, and communication capabilities. These sensors are capable of detecting special events called "danger" (e. g. temperature, biochemical agents, etc. ) that are above a particular threshold. The sensors that have triggered the special events are considered to be obstacles.
3. 2 Sensor Network Implemented in BreveThe sensor network mentioned in the previous section was modeled in Breve using patches. The patch class in the Steve language was utilised for this purpose. In this paper, the patches were equally distributed over the entire region. But efficiency could have been improved by placing the patches in a certain fixed pattern to minimize the number of patches and thus maximizing the safety of a vehicle navigating through the area infested with "danger" (obstacles). In [6] the sensor deployment problem in the context of uncertainty in sensor locations for airdropping situations was considered.
Sensor deployment in such scenarios is inherently non-deterministic and there is a certain degree of randomness associated with the location of a sensor in the sensor field. In a sensor network the sensors would sense the special events electronically. This is simulated in Breve by placing light obstacles over the patches. The patches are capable of sensing obstacles placed on them.
By getting the location of the light object and by finding the patch present at that location we are able to determine the patches which have obstacles (or "danger") and patches which are safe.According to the principle of Cell Decomposition the entire world is decomposed into equal sized patches. The Patch and PatchGrid classes in Steve were used for this purpose. These patches provide a sense of location. That is, given a location it is possible to obtain the patch object residing in that location.
The patches were created using patches = (new PatchGrid init-at location (0,0. 75,0) with-patch-size (5, 0. 1, 5) with-x-count X_SIZE with-y-count Y_SIZE with-z-count 6 with-patch-class "LifePatch"). By changing the x and z values it was possible to create patches of different dimension e.g.
, 4x4, 6x6 etc. A 32x32 patch would fill the entire work space created in this simulation. 4. 2 Light Objects Modelled as Obstacles The light objects which are part of the sample Braitenberg class in Breve were used as obstacles. Light objects are mobile objects and can be moved around during the simulation, which provides us with dynamic obstacles i.
e. obstacles that would change their position with time. In this work it was possible to infer a safe path avoiding the obstacles dynamically by checking for changes in the position of the obstacles at every iteration time step.If it were not for computational limitations, this type of replanning from scratch would be the ideal, since it guarantees optimal plan generation and execution given all known information at the time it is acquired. The use of the D algorithm ( Dynamic A) could lessen the computational work by producing an initial plan based on known and assumed information, and then incrementally update the plan as new information is discovered.
The implementation of the D algorithm is left as future work.
