Unmanned vehicles: autonomous and on-target

June 1, 2005
For the most part, “unmanned vehicle” is, today, something of a misnomer.

By Roger Joel
Vice President, Sales & Marketing
Octec Limited, part of Radstone Embedded Computing

For the most part, “unmanned vehicle” is, today, something of a misnomer. While the “man” in question may not physically be in the vehicle, there are few so-called unmanned vehicles currently deployed that do not largely depend on human intervention for the successful completion of a mission.

Roger Joel
Click here to enlarge image

“Autonomy” has long been the goal of unmanned vehicle development-but autonomy has, to some extent, been redefined according to the abilities and limitations of the technology available at the time. For an unmanned vehicle to be truly autonomous it should, of course, be capable of completing a mission entirely unaided by external influences.

While semiautonomous unmanned vehicles meet many of the goals of truly autonomous vehicles-removing personnel from risk and allowing operations in environments that would be too hazardous for direct human involvement-there are others that are not met, including substantially reduced reliance on operator/vehicle communications and a reduction in the total manpower deployed.

In the first case, there is widespread recognition of the problems inherent in absolute reliance on communications between man and vehicle-the possibility of interception and/or jamming, the ease with which the communicating vehicle can be detected, or loss of signal due to equipment failure or geographic/climatic interruptions. Beyond that, there is equally widespread recognition that much of what is transmitted from an unmanned vehicle to its operator is irrelevant, providing neither information nor insight-the limited available bandwidth is wasted. All of the transmission has to be viewed, however, in order to glean the small amount of valuable information that’s contained within it.

With regard to reducing the manpower deployed, it is still the case that many unmanned vehicles require two operators at the ground station-one to “fly” the vehicle and the other to manage the payload operations. While increasing flight-path autonomy of UAVs (a supervisory presence still being required) has decreased reliance on the former, the latter remains a labor-intensive process.

Separating functions

For all of the operational goals of unmanned vehicles to be met, 100 percent of the functionality of the driver or pilot would need to be replaced. Two key elements would be required to do this: the ability to “see” and interpret video images and the ability to act on the information obtained.

Bearing in mind that the integrity and safety of the mission must not be compromised, a possible system architecture is outlined in the figure on p. 11. Note that payload processing would be separated from the flight-management processor. This separation would remove the payload-processing function from flight safety-critical issues. The payload-management processor would not be empowered to make mission decisions. Rather, based on the data from the sensors, it would make a request of the flight-management processor (FMP) to deviate from the original plan. The flight-management processor would consider the request in light of the higher-level objectives of the mission’s overall flight safety and the flight envelope of the vehicle and either authorize or deny the request.

By splitting the functionality within the payload-management processor a growth path is provided for the fusion and subsequent processing of data from other vehicle payloads such as synthetic aperture radar. Those image-processing functions available today include target detection and tracking, electronic image stabilization, image enhancement, image matching, image fusion, and image mosaicing.

The architecture is designed to achieve three objectives: a reduction in system complexity; the flexibility to divide development, enabling progress in each area to be made in parallel; and, most important, to maximize mission safety.

Implicit in the architecture, however, is the ability for the vehicle to “see” its surroundings-via its electro-optical payload-and to take appropriate action as recommended by the payload-management processor but authorized by the flight-management processor.

System architecture that separates the payload-management processor from the flight-management processor, would allow the UAV to “see” and interpret video images, and act on that information.
Click here to enlarge image

In theory, the technology to allow such a system to be developed has existed for some time. There were, for example, numerous image-processing algorithms in the public domain. However, the problem with these was that they were never designed to run in real time. While a one- or two-second image-processing delay is acceptable in, for example, face recognition in security applications, it is patently unacceptable in a combat zone. However, Octec-part of Radstone Embedded Computing-has developed new algorithms and optimized existing image-processing techniques so as to deliver actionable data in real time. Of course, overall suitability for unmanned-vehicle applications still requires a solution that is compact, lightweight, and consumes minimal power. It is through the combination of optimal algorithms and special rugged hardware-based upon a processor/field-programable gate-array (FPGA) combination-that Octec can now offer a solution to unmanned-vehicle payload management.

Typical of the type of processor that might be used as a payload-management processor is the Octec ADEPT104, which combines a high-density FPGA with a Motorola PowerPC. The ADEPT104, which is already operating with a number of customers, is the first example of a new family of very compact image-processing modules.

Using the gift of sight

So what might an unmanned vehicle with the gift of sight be able to achieve? A typical mission might involve reconnaissance to ascertain the level of damage to a dam generating complex resulting from a prior bombing raid. An image of the dam area would be loaded into the payload-management processor prior to UAV takeoff, with the specific area of interest marked. This image may have been produced from a previous mission or from satellite reconnaissance. Geographical information relating to its location would be loaded as data. As the original image was probably taken at a different altitude and angle from the planned approach path of the UAV, sophisticated warping algorithms would enable the reference image to be directly compared with the live video. Once the target is “matched” and acquired, images of the appropriate resolution would be captured and stored. Given that the requirement was for the vehicle to maintain “radio silence” in the mission area, the UAV would automatically navigate to a downlink point.

In such a scenario, the FMP would navigate the UAV to the target location; the PMP would determine the direction and height of approach based on geographical, sensor, and meteorological data; the PMP would assess the captured images for adequacy and then advise the FMP that the mission is complete; the FMP would navigate to the downlink position, where the PMP would optimize the data for transmission-minimizing data redundancy (perhaps by mosaicing the image, or by fusing IR and TV imagery), thus minimizing bandwidth usage while providing only the required information to ground personnel-and initiate the downlink. Note also that processing/reducing the data onboard minimizes vehicle-based storage requirements.

Perhaps the single biggest challenge-both to truly autonomous unmanned vehicle designers and to the widespread military deployment of unmanned vehicles-is obstacle recognition and avoidance. The relatively slow speeds of land and underwater vehicles pose less of a threat, but the goal for UAVs is to demonstrate that unmanned vehicles can operate with a level of safety comparable to that of manned vehicles. That means, for example, that they will need to be able to prove that they can “see” as far as the human eye in similar conditions-a distance of around three miles-and take the action to avoid collision both with other air vehicles and the ground.

True autonomy of unmanned vehicles is not yet with us. Unmanned vehicles are still, for the most part, manned in at least some way. However, the rewards of true autonomy are sufficiently great, and the technology moving forward sufficiently quickly, that it’s difficult to believe that the current situation will last for long.

Voice your opinion!

To join the conversation, and become an exclusive member of Military Aerospace, create an account today!