Figure 1:  REEF Visualization Laboratory:  (a)  Computer cluster for generating up to 10 camera images within the virtual environment;  (b)  Plasma TVs displaying computer-generated images from the UF campus virtual environment.

The REEF Visualization Laboratory, funded via an AFOSR FY-04 DURIP grant, is a key component in the REEF’s experimental capabilities for studying vision-based control of autonomous vehicles.  The lab, shown in Figure 1, is composed of a cluster of 11 computers that are able to generate 10 different camera views within a virtual environment.  Multigen-Paradigm’s Vega Prime 2.0.2 software is used to generate each virtual camera image given the position and orientation of the camera within the environment.  Four images can be viewed at one time on 50-inch plasma screen displays.  The lab is capable of supporting hardware-in-the-loop simulations which enables researchers to study the performance of image processing and vision-based control algorithms using the same hardware that will be used in flight.  For example, in Figure 1(b), a camera mounted on a small UAV is recording images from the plasma TV on the far right. The REEF Visualization Lab currently has 2 virtual databases:  a UF campus database developed by the UF Digital Worlds Institute and a database of the Military Operations in Urban Terrain (MOUT) site at Ft. Benning developed by Quantum 3D. Both of these databases, which are shown in Figure 2, provide useful virtual environments for studying autonomous flight in urban environments.

Figure 3 illustrates the general framework for closed-loop simulation of vision-based control algorithms in the REEF Visualization Lab.  In this figure, a vehicle dynamic model determines the position and orientation of the UAV, which in turn specifies the pose (i.e., the position and orientation) of a camera mounted on the vehicle.  Given the pose of the camera, the image generation software creates and displays a camera image within the virtual environment.  A flight camera then records the virtual camera image from the plasma TV display.  The recorded image is passed to an image processing algorithm which extracts useful information from the image.  This information is then used by a vision-based control algorithm to generate the appropriate control commands.  For example, the control objective might be to navigate the vehicle to a goal location while avoiding obstacles in its path.  The control commands are passed as input to the vehicle dynamic model which then determines the next position and orientation of the vehicle.

Figure 2:  Virtual environments for vision-based control research.

Figure 3:  Vision-based control simulation framework.