Hudson Research Home Page

         Artificial Intelligence

Hudson Research Inc has recently entered the field of Artificial Intelligence (AI) and Unmanned Autonomous Vehicles (UAV) as a result of requirements associated with our Space Shuttle Program (G-301). The Get-Away-Special (GAS) program imposes many unique requirements on the system controller. In this program, astronaut involvement is minimal, primarily limited to turning the instrument on at some random point shortly after launch and arrival in orbit and turning the instrument off towards the end of the mission. The G-301 Hyperspectral Imaging Spectroscope (HIS) experiment is one of the most complex experiments to be carried under this program. The experiment utilizes an advanced uncooled Infra-red hyperspectral imaging sensor controlled by an artificial intelligence program with an embedded Global Positioning System (SatNav; GPS) capability. The mission specifications call for the instrument to have continuous awareness of its geographic position in order to determine when to turn the imaging system on to acquire images. The host must be able to store a large number of potential targets and camera setup information associated with them and be able to compare its position to the target database. Once it has been determined that a target is in acquisition range, the host then transmits instrument setup data specific to that target to the camera controller.

The situation is complicated by two issues:

	   (1)  Shuttle orientation relative to Earth
	   (2)  Cloud cover over the target

This problem has been resolved by use of an artificial intelligence program developed in-house specifically for this application. In order for acquisition to occur, three positive votes must be cast in the AI sub-routine. They are position and two factors derived from scene content.

Target Determination (TD) Program: This is the highest level program. It receives data from several sources and based on a set of rules determines whether the image acquisition sequence should be activated and what the specific camera settings should be. The TD program receives data from the Navigation program (NAV) and compares it to a list contained in the Target database. Each target is listed as a set of geographic coordinates. When a current value from the NAV program falls within the values of a predetermined target region, then a positive vote is cast in the Polling sub-routine. This causes the frame grabber to acquire a frame from the imager. This frame is then analyzed in the Polling sub-routine described below.

Polling Sub-Routine: The thermal content of the frame is examined for determination of the percentage of deep space visible to the imager. Determination is made on the basis of the average temperature of the scene. The average temperature of a deep space image is almost always negative whereas the average temperature of an earth view falls within a narrow band of positive values. If the average temperature falls within this range, a second positive vote is cast to the DM subroutine.

A second analysis is performed on the data to determine if there are clouds in the field of view. This analysis is a combination of water vapor content in the image field in proportion to the percentage of whiteness. If the cloud analysis is acceptable then a third positive vote is cast to the DM routine. The accumulation of three positive votes causes the activation of a targeting sub-routine. The selected target profile is read from the target database and those values are fed to the camera interface. The camera is activated for a pre-determined period and scan configuration, and time, cloud cover data, position data from the GPS are recorded in the interframe intervals onto the hard disks.

Navigation (NAV) Program: The navigation program receives periodic position and time data from the GPS receiver. It uses this data combined with a priori knowledge of the predicted orbital track to calculate a continuously updated position. The predicted values are used to extend the position in the absence of GPS signals. This is an important aspect since the GPS satellites will not be continuously visible to the antenna. It provides output to the Target Determination Program. When GPS data is available the fix is updated.

Target & Rules Database: The Target Database contains a listing of the exact geographical coordinates of each a target, the acceptable position guard band, and camera set-up information. The Rules Database is realized as a set of batch files. It contains a listing of all the decision making rules.

Figure 1   G-301 Artificial Intelligence Architecture

The interrelationship of the hardware and software components of this system are shown in Figure 1 . For further information or if you have an application you would like to have us analyze and develop, CONTACT US

        
                     
             Copyright  2006  Hudson Research IncNew York