Unmanned Ground Vehicle Reconnaissance
Increased world peacekeeping activity has increased the need for reconnaissance, frequent surveillance and even constant monitoring, with reduced public acceptance of danger to peacekeepers. This may already drive the market for advanced land vehicle reconnaissance technologies, such as those found in the Canadian Light Armoured Vehicle Reconnaissance (LAV Recce) and other programs. It has been commonly speculated that future battlefield and other hot-spot situations will see increased use of telerobotic or even autonomous vehicle systems for remote information gathering and monitoring. While universities, private research institutes, and even the Defence Research Establishments have been pursuing Unmanned Ground Vehicle (UGV) developments, UGV Reconnaissance (UGV Recce) will require surveillance systems, including sensors, signal processing, communications and man-machine interfaces, from those who are now developing such technologies for the LAV market.
This paper will discuss a number of elements and technology challenges that may be important to future UGV Recce systems, drawing on a large quantity of work that has already been done in separate areas of autonomous vehicles (robotics) and surveillance sensor technology. An attempt has been made to cover a fairly broad collection of technologies, both to consider missing background, and perhaps to point to areas of particular interest to the intended reader. The main purpose of this discussion paper is simply to begin exploring an area, UGV Recce, that many may already be involved with through work on related technologies, but which seems rather unexplored as a whole.
The next section introduces UGV Recce, with attention to the move toward unmanned vehicles and the characteristics of autonomy. Section 3 introduces and discusses autonomous vehicles, highlighting aspects of mobility, power and navigation, and how technology in these areas may relate to autonomous vehicular reconnaissance. Section 4 discusses reconnaissance sensor technologies, including sensor deployment, and offers for consideration a few characteristics and possibilities of command, control and communications in future UGV Recce systems. Section 5 concludes this work by summarizing what has been discussed.
2. BEYOND LINE-OF-SIGHT RECONNAISSANCE TO UGV RECCE
Beyond line-of-sight (LOS) reconnaissance requires the ability to acquire information from an area not directly observable from the current location. To go beyond the current LOS one could go on foot or by vehicle. Safety issues aside, one would choose to go by vehicle to be able to transport more surveillance, communications or other equipment more quickly and easily, and to be able to return or move to some other position, again more quickly and easily. Admittedly, if little equipment is needed, and one only needs to survey an area that is literally ‘just over the hill’, it may be better to do it on foot.
It is pointed out in  that unattended ground sensing is seen to be a major need by the Canadian Forces for beyond LOS surveillance. Admittedly, this need does not explicitly require vehicles or autonomy. However, surveillance installations in hostile territory require initial movement to, and possibly within, that territory. Once in place, the greater the autonomy of the remote system, the less direct attention (monitoring/servicing) required by personnel. Finally, it is highly desirable that any required removal and/or relocation of surveillance equipment does not require risk of exposure of personnel.
Sophisticated land reconnaissance is performed by sending vehicles to remote areas to be investigated. Questions that may be asked once the choice to use a vehicle has been made include: Should the vehicle be air-borne or land based, should it be manned or unmanned, and (in this work) how might autonomy come into play? The following two subsections briefly consider and propose answers to these questions.
2.1 Unmanned Vehicles
Aerial land reconnaissance facilitates rather fast, deep penetration into remote areas (ex. search and rescue planes/helicopters). Low-level manned reconnaissance (i.e. slow) flights can however be overly hazardous when investigating hostile areas. Considering a remedy to this problem, we note that an area that is undergoing active development is that of air-borne Remotely Piloted Vehicle (RPV) reconnaissance . Without attempting to balance an argument, it is immediately noted that unlike powered air vehicles, ground vehicles can operate for much longer periods (days or conceivably weeks) without servicing, while remaining completely undetected. It would seem to be rather difficult to sneak-up, cross front lines, perform covert surveillance for an extended period, and finally return undetected, using an air-borne RPV.
The Light Armoured Vehicle Reconnaissance (LAV Recce) system  is representative of the state-of-the-art in land vehicle reconnaissance. The basic standalone system consists of a manned, 8-wheeled Armoured Personnel Carrier (APC), that has been outfitted with an extensive suite of surveillance and targeting sensors, operator’s console and communications equipment. LOS sensors, including radar and thermal imaging devices, can be extended on a mast to about 10 m above ground. In addition, other non-standard surveillance equipment or techniques could also implemented on short notice, since additional equipment would likely not affect the vehicle’s ability to maneuver and bring sensors to bear.
Considering overhead to manned ground vehicle reconnaissance, note that in addition to armour, the LAV Recce APC has a turret-mounted canon and machine guns, assorted protection and survival equipment, and crew supplies for extended missions. In fact, perhaps the greatest amount of effort and cost in developing, procuring and supporting the LAV Recce system is not the surveillance technology, but the bringing of the operators to the surveillance site. Unmanned ground vehicle reconnaissance, UGV Recce, on the other hand, would not require on-site crew, heavy armour, turret, guns, ammunition, fire-control system, shrapnel containment and flame suppression, substantial heating and air conditioning, survival gear or food supplies. Also, there would be much less system weight and more free space, reducing the vehicle profile and leaving more room for fuel, backup power, surveillance sensors and communications systems. Finally, the UGV Recce system could operate for long periods in the field without service or supply, and may be left if need be, without consideration for personnel.
Assuming that one has an unmanned vehicle for reconnaissance, there are different levels of autonomy that might be desired of the system. For example, a highly autonomous (robotic) UGV may be instructed to move about a range while performing relatively high-level monitoring, doing further surveillance and reporting when necessary. A less independent system might be described as semi-autonomous, where the vehicle is remotely guided or piloted to some degree. The UGV might automatically track a path and avoid obstacles without supervision, but it might not ‘know’ a final destination, goal or mission, and specific path segments along the route might be chosen and actively supervised by the operator. In the least autonomous mode, an operator might remotely drive the vehicle, which would seem to be reminiscent of RPV reconnaissance applications. This may represent the entry level for development of UGV Recce systems. The labeling of a system as fully autonomous or semi-autonomous, or somewhere in between, seems to depend on how much teleoperation (real-time remote control) the system requires.
It is noted that two broad areas of autonomy appear to be available for pursuit; that of vehicular activity and that of surveillance. This work will not attempt analysis of an appropriate balance of autonomy between these two areas, or to determine if there may be more levels or regimes of autonomy to explore. (Admittedly, such analyses might be very interesting and useful.) Suffice it to say that there are various reasonable scenarios or configurations for the UGV Recce system concept, and that it is also interesting and useful to simply pick applications and system scenarios at hand. Therefore, it is noted that two items recently highlighted in  as requiring unattended activity are 1) the ability to differentiate between vehicles and personnel, and 2) 24-hour, all-weather monitoring of perimeters, and early warning;
Gross differentiation between vehicles and personnel would seem to be an activity that could indeed be done automatically. In fact it should not require highly speculative future technologies. It is noted that various automatic target detection  and classification  capabilities are already under investigation.
Requirements for all weather, 24-hour perimeter monitoring and early warning would seem to match the inherent capabilities of an unmanned surveillance vehicle. It is noted in  that the LAV Recce system , on which the vehicle and sensor capabilities for this UGV Recce work are heavily based, in fact already partly meets these very needs. The required autonomous monitoring and reporting activities could be seen as challenging extensions of developments in commercial security robot technology.
It is again noted that the above are simply recent applications development examples that were close at hand. However, it seems noteworthy that they are already considered to be needed, appear achievable, and would lend themselves nicely to desired initial UGV Recce applications.
3. AUTONOMOUS VEHICLES
The development of new autonomous vehicle concepts and related technologies has increased dramatically over the last decade or so. Greatly increased high-speed computational capabilities have supported relatively high-level autonomous road vehicle applications development, an example of which has been described in . Such work has obvious applications to future intelligent vehicle/highway systems development, which is documented in  and elsewhere. Complementary to this, the development and proliferation of inexpensive microcontrollers, as well as novel sensors, has supported and encouraged new low-level architectures for smaller autonomous mobile robots [8,9].
The discussion here will be limited to high-level considerations of autonomous vehicles, with occasional connections made to lower-level issues where it is believed to be necessary or particularly interesting to do so. Therefore, it is noted that primary considerations for an autonomous vehicle include its physical mobility, the availability of power, and the ability to guide and control its own actions, referred to here as navigation. Points of interest in these areas are discussed below, with the intent that they offer useful background and perhaps spark further interest in UGV Recce systems-level considerations of autonomy.
Mobility may be defined as the cumulative measure of agility (acceleration/deceleration), maneuverability (turning capability), gradability (slope climbing/descending), trafficability (soft soil, ditch/boulder crossing) and ride quality (shock/vibration isolation). A vehicle’s required mobility may be based primarily on only a subset of the above factors. Given two different vehicles designed with similar capabilities in the above, the vehicle that can move more rapidly, through terrain that comprehensively tests the corresponding factors, would be considered to have the higher mobility.
For a manned LAV Recce system, high speed travel with high performance in all five of the above areas would seem to be desirous (required), whereas for future UGV Recce, perhaps agility, ride quality and top speed might be sacrificed for higher performance in the remaining mobility factors. This is perhaps practically realized through a few general vehicle parameters such as size, weight, traction and ground clearance, and by a collection of specific capabilities such as vertical stepping, climbing and ditch crossing. Certainly typical road vehicle performance parameters such as acceleration, speed and deceleration distance could be of significantly less importance for typical UGV Recce off-road activities.
The mobility of a vehicle generally increases with its size and number of degrees of freedom in its frame. Generally, a vehicle can overcome obstacles of a size comparable to a characteristic dimension (say wheel diameter for a rolling vehicle), with variation in capability determined by the vehicle’s structural variability (degrees-of-freedom). For instance, a three-axle, highly articulated vehicle (i.e. single-axle front, centre and rear cabs, with an actuated joint between cab pairs), might climb over a rock about as high as a wheel diameter, and cross a ditch about a half of a vehicle length across. If articulations are removed, the maximum obstacle size might reduce to say less than half a wheel diameter, and if two axles are used (like a road car), maximum ditch crossing width could reduce to less than a wheel diameter. Articulated vehicles are somewhat popular in planetary exploration endeavors. A proposed three-segment, multi-wheeled planetary rover, described in , has 1 m diameter wheels, and is indeed expected to climb over 1 m diameter obstacles. On a much smaller scale, the 11.5 kg Sojourner rover, intended for the Mars PathFinder mission , has 13 cm diameter wheels, and will also apparently overcome 13 cm obstacles.
This work has so far assumed rather conventional notions of mobility. Specifically, proposed future UGV’s and their characteristics have been assumed to be based on vehicles that roll. This class of vehicles includes road cars, motorcycles, snowmobiles, tanks, and virtually all other land vehicles used today. A second class, those that step rather than roll, are not conceptually new. Land animals, including elephants, horses, cheetahs and humans, have inspired efforts to develop walking vehicles. Although progress has indeed been made , and certainly continues, the problem of dynamic stability is a very challenging one.
As an alternative to dynamically stabile legged motion, development programs pursuing insect-like mechanisms have progressed further in a practical sense, by using lighter structures capable of static stability. Essentially, if power, computational or otherwise, runs low, the vehicle doesn’t fall over. One example is the six legged robot referred to as Ambler . Conceived for planetary exploration, it can manage step heights of 1.5 m, but has a top speed less than 0.1 km/h. Recent field work involving a similarly capable eight legged stepper, named Dante II , saw it climb into (and out of) the mouth of a volcano in an effort to gather data too dangerous to gather on foot. Traversal of the inner wall at the mouth of a volcano would likely not be safely accomplished using a typical rolling vehicle. In fact, legged locomotion would seem to allow access to virtually all land on the earth, while a U.S. Army study, quoted in , suggests that about 50% of the earth land mass is inaccessible to wheeled or tracked vehicles.
The following terrain traversal characteristics are described mainly in terms of advantages of legged vehicles over rolling (wheeled or tracked) vehicles. Although rolling vehicles would be the natural extension of current LAV Recce systems to UGV Recce systems, the legged-versus-rolling framework may help highlight considerations and desired capabilities for all land vehicles.
maneuverability/trafficability A legged vehicle is able to move in three dimensions, while a wheeled or tracked vehicle must move in two . A legged vehicle is able to utilize isolated footholds .
speed Average speed in rough terrain is noted to be about 5-8 km/h for a wheeled vehicle, 8-16 km/h for a tracked vehicle, and up to about 55 km/h for an animal . In addition, some rolling vehicles achieve animal speeds, but only with very high power consumption and damage to the environment . (Admittedly, many current walking vehicles achieve speeds of only about 0.1 km/h, as noted earlier.)
efficiency De-coupling (or otherwise minimizing) vehicle interaction with the terrain avoids unnecessary work  (ex. a car suspension system). It is noted that a leg/foot tends to step on the ground, while a wheel often ploughs and is continually digging itself out of its own rut . Superiority of one system over another may still vary however, depending on specific foot and wheel interaction with each soil type.
vehicle vibration/shock Passive suspension, often found on a rolling vehicle, leaves that vehicle subject to unpredictable dynamics. Adaptive legged vehicles are inherently actively stabilized, and can isolate the vehicle from terrain irregularities , especially during fast motion. Additionally, during step height changes, wheeled vehicles may be forced to sustain substantial shocks, whereas legs may step up/down.
terrain impact Rolling vehicles cause surface abrasion and even ploughing, and tracked vehicles typically tear-up a surface using grousers. In peaceful surveillance applications, or during movement between operational theaters, it is highly desirable to avoid surface damage. During covert reconnaissance, this may be doubly desirable. Although both legged and wheeled vehicles may cause surface indentation, on reasonably firm soil the legged stepping action should leave less identifiable traces, and in difficult terrain, obstacles that can be stepped past must often be rolled over or through.
The primary power source required for a vehicle is strongly dictated by desired drive characteristics. A large vehicle like a manned LAV, that moves fairly quickly, requires the high specific power of a fuel engine. If one can use a lighter vehicle and/or accept lower speed, other alternatives, most notably electric or fuel cell technologies, become viable. We will not examine primary source, generation or conversion technologies, however we will try to estimate what the power requirements might be for future UGV Recce systems.
To put power levels into perspective, we can examine a few specifications from across a range of vehicles. Considering engine power and vehicle weight, we note: Abrams Tank: 1100 kW, 55 t (20 W/kg), Bradley APC: 450 kW, 30 t (15 W/kg), LAV Recce: 200 kW, 13 t (15 W/kg), road car 100 kW, 1000 kg (100 W/kg), Landscaping Tractor 15 kW, 600 kg (25 W/kg). A future reconnaissance UGV, with a frame of a size and weight similar to one of the above examples, would likely require comparatively less power to address its performance requirements; there are no heavy armaments/implements and no people to quickly move about. As an aside, planetary exploration might be looked to as an extreme application of unmanned low power rovers. The 11.5 kg rover proposed for the Mars PathFinder mission  uses a 16 W peak solar-electric array (1.4 W/kg), while on the other end of the weight scale, a 1000 kg planetary explorer, proposed in  and similarly elsewhere, might use a 500 W radio isotope thermal generator (0.5 W/kg).
Generation of motive power is a less than covert operation, due to acoustic and/or thermal emissions. It seems then that the above power ranges should represent extreme upper boundaries for continuous power for surveillance and other equipment. In fact, since part of an UGV Recce mission is likely to consist of extended, isolated surveillance, energy resources should also be conserved, pushing still harder for designs that do not require continuous high power. (Intermittent power requirements could be allowed to exceed the maximum power available from the primary source, or intermediate conversion system, using a large battery or other storage buffer.) To facilitate a systems level consideration of power, one could note some of the following power requirements that may impact near future UGV Recce:
computers If all high-level vehicle and surveillance related activities could be handled by two high-performance computers (2 ´ 250 W), and a two-channel high-performance signal processing system with DSP cards, I/O and A/D peripherals (250 W), we might estimate a continuous power of say 750 W.
communications Assuming that normal continuous reconnaissance activities use low data-rate networked communications, perhaps only a few Watts (maximum) would be required to sustain a radio communications transceiver. At high levels of direct-to-operator data transmission, for example in close-range detailed surveillance, perhaps 50 W would be required for sustained transceiver operation.
sensors At the core of current and proposed LAV Recce systems are high resolution thermal and visible light imagers, and some form of scanning radar unit. In addition to pan-tilt and stabilization systems, mechanical drive is required for most land reconnaissance radar systems, and thermal imagers typically require some form of active cooling. With these and a few other specific systems engaged, one can expect to require say 500 W for normal operation. Additional sensors would of course add to this requirement.
deployment While sensor (head) deployment may not require continuous power, initial deployment and intermittent (but frequent) adjustment could outweigh all other surveillance power requirements. Assuming the UGV Recce system is used in a direct-view telepresence mode, one would expect that the distant operator would desire deployment control and speed of the order of that required for manned vehicle reconnaissance. Experience with mast-mounted deployment for LAV Recce suggests elevation of just the few pieces of equipment noted above can require power of the order of 2.5 kW with a practical telescoping mast. Calculations have shown that this requirement is almost an order of magnitude higher than should be expected for a loss-less drive system. One might therefore assume a future improved requirement of say 500 W - 2500 W for intermittent elevation of a sensor suite.
Summing the above equipment power requirements, one can estimate that surveillance activities might require continuous power of about 1.3 kW, with intermittent peaks in the range of 1.8 kW - 3.8 kW. While intermittent peaks may not greatly affect energy reserves, it may alternatively be possible to benefit from intermittent or extended drops in power requirements. During low apparent target activity it may be reasonable to power-down sensors for some period, or to cycle them off and on in some optimum (perhaps random) manner. This energy conservation tactic could be commanded during teleoperation, or executed by a highly autonomous system during isolated auto-detection, tracking and early warning activities.
Although power consumption may tend to fall for any given technology, and with introduction of new operating techniques, there are likely to be new technologies (and tasks) to add to a future UGV Recce system, again pushing power requirements upwards. In the past, high mobility and heavy (LAV) systems would seem to have dictated the primary power source design, with other on-board equipment needs being secondary (and less). In the future, much less power may be required for UGV’s, while the same or more power may be required to use the current sensors with new ones that may become available. On-board equipment may dictate future power requirements, with drive power being secondary.
In the context of autonomous land vehicle operations, navigation might best be described as encompassing the two functions of motion planning and motion control. This very broad definition attempts to include the many elements required to move a vehicle intelligently. Table 1, adapted from consideration of autonomous planetary vehicle navigation , illustrates how these elements might be broken out to reveal various operations and parameters. Activities towards the left in the table are at a higher command level than those toward the right, and are where an operator would tend to interface with a sophisticated UGV. Low-level interface points, toward the right in the table, would likely be required for a near-future UGV Recce system, but would become decreasingly necessary with progress in robotics, cybernetics, and other areas of artificial intelligence. To place this scheme into perspective, it is noted that somewhat mid-level (left to right) elements, such as use of a magnetic compass, the Global Positioning System (GPS), and dead reckoning, are central elements of current manned LAV Recce systems .
Table 1: Autonomous Navigation Considerations for UGV Reconnaissance
Navigation schemes for autonomous or semi-autonomous vehicles are frequently arranged such that higher-level path-planning activities specify paths to be taken, with no concern for the path traversal (motion control). A desired path is defined in some coordinate system, and the motion control system steers the vehicle along the path, using some system of reference to indicate its error in doing so. Knowing the path in absolute world coordinates, the motion controller could use the GPS, Inertial Navigation (IN) and even odometry to determine its relative error in following a given path. A fairly straightforward steering algorithm can be used to track the target path.
Alternatively, an UGV might move toward a goal without following a pre-planned path. If a specific goal is chosen at some distant point, the UGV would simply head toward the goal (likely measuring its progress using the GPS, IN and odometry as before) in much the same way as a human might. Travel occurs then with little regard for the path taken, and obstacles are simply dealt with (circumvented) as encountered. With little pre-planning, such techniques are often referred to as reactive or reflexive motion control.
Each of the above path tracking approaches involves a somewhat straightforward method of approaching a goal; the first by staying on a path, and the second by keeping a heading in the direction of a goal point. Each technique also requires a method for avoiding and circumventing minor obstacles, much as a human steps to the side of a hole or rock, and deviates from his intended direction for short periods to pass an obstacle. These obstacle avoidance techniques can simply involve slight modification of the progressing motion using some type of bias signal to the steering controller. This signal can be generated from medium to low-level sensing, such as identification of object position from image analysis, laser scanning, infrared and acoustic proximity, and even tactile sensing.
Finally, it is speculated that complementary to the notion that planning and high-level control operations might be augmented by operator input, the navigation system may incorporate various aspects of behavioral control schemes . For instance, if the autonomous UGV is stopped to use its radar for surveillance, a minor detection of motion might cause deployment or pointing of other sensors in the general direction of the detection. This would then not be a hard system trigger for any particular action or operating mode, but simply the addition of a parameter that influences the general scanning characteristics (attention) of the system, which may still scan other directions and move from location to location. Further signals by ‘enough’ other sensors might extend the UGV’s stay in one place, allowing for more detailed sensing, analysis and operator notification. Alternatively, the increased probability of a threat might discourage the system from using certain types of active sensors, encourage some sort of silent running, or even encourage a move toward more cover.
4. RECCONAISSANCE TECHNOLOGY
Beyond line-of-sight reconnaissance, discussed earlier, may be taken in a broad sense to imply the gathering of many types of information about an area beyond ones current location. These types of information might extend from radar imagery, to sound, air and soil samples. Gathering data from such diverse sources not only requires various sensors, but brings to the forefront the consideration of deployment methods and related technologies. Also, remote deployment, sensing and other operations seem to mandate some consideration of other systems issues. The following subsections briefly discuss possible UGV Recce sensors and their deployment, and consider a few possibilities for command, control and communications characteristics for the UGV Recce system operation.
A very common approach to sensor system design, somewhat akin to biological systems, is to choose a suite consisting of sensors that complement each other. This may be contrasted with choosing a single high performance sensor, or several sensors of a similar type. The complementary approach not only ensures that a single failure stands little chance of cutting-off the information gathering ability, but also that few surveillance targets would be able to elude detection of all sensor types. In addition, a single high performance sensor can be prohibitively expensive, and may not be mature enough for field applications.
Sensor fusion, which may be considered to be a similar but perhaps lower level process from data fusion, might be described as an elegant use of complementary sensors and their outputs, with some form of fusion process taking place close to the source. The resulting output can simplify the burden on higher level systems (including operators), as well as simplifying systems design. To accomplish this, one could choose complementary devices at the beginning of the systems development. For example, in the case of the motion control system, infrared proximity and tactile sensors, collocated on the vehicle, might combine their outputs, such that the higher level system receives only a single three-state output: low (clear), close proximity, and contact. In later development, this one signal source can be used in obstacle avoidance processing, in place of two. Fusion processes can also be extended to higher-level separated surveillance sensors, or to entirely different types of data. There are numerous teleoperated systems (RPV’s) that have required developers to come to grip with the remote vehicle sensor and data fusion challenge. Recent consideration of data fusion issues relating to teleoperated remote landmine detection are described in .
The following points describe a number of sensor types and their variants, in an attempt to cover as wide a detection range as possible. A number of the sensors listed may seem somewhat esoteric or simply impractical, however the choice to list these technologies stems from a desire to encourage future systems debate, while at the same time keeping in mind what would be possible for future UGV Recce.
Visible Light Imaging The LAV Recce system  has a high-resolution visible light camera (monochrome) with zoom capability. The camera is mounted on a pan-tilt platform, which in turn is elevated above the vehicle using a powered free-standing mast. Initial UGV Recce applications could certainly utilize a similar system for teleoperated activities, however deployment and pointing control may pose a challenge for highly automated systems at this time. Fixed-focus roof-mounted systems would likely represent a good entry level approach for automation.
Thermal Imaging The LAV Recce system  has an 8-12 um wavelength thermal imager (TI). It is mounted parallel to the visible light camera, on the same pan-tilt platform mentioned above. A thermal imager is a very likely candidate sensing technology for UGV Recce applications. In fact, for automated target detection (hot vehicles for instance) this may be the line-of-sight sensor of choice.
Intensified Light Imaging The commander’s sight on the LAV Recce system  contains an intensified light imager for very low light-level viewing. These systems are also now becoming a common tool for the individual soldier, in the form of night-vision goggles. Although it is currently a valuable alternative to using much more expensive thermal imagers, thermal imager costs are currently dropping, and any savings over TI’s would not likely represent much of the total cost for initial UGV Recce production systems.
Stereo Vision Stereo vision is now frequently applied in robotics developments. It operates under the same principal as mammalian two-eye systems, using parallax of two viewing angles to give depth information. Such a system would lend itself well perhaps to advanced UGV telepresence applications for surveillance, in addition to image (scene) analysis required for obstacle avoidance techniques noted earlier.
Laser Range Finding The LAV Recce system  uses an eye-safe laser range finder that is mounted and aligned parallel to the visible-light and thermal imagers, on their common pan-tilt cradle. In UGV Recce applications, such a device would likely be used during teleoperated surveillance. However, difficulties in aiming and interpretation of results may require modification of typical systems for autonomous operation.
Scanning Range Finding These systems, which are scanning versions of the above, with surface profiling capabilities, are typically used in mobile robot applications to locate and define objects at close range. Such devices may be of value in highly autonomous UGV Recce navigation systems.
Radar The LAV Recce system uses a 17 GHz sector scanning Doppler radar system, capable of detecting a moving vehicle at a range from 50 m to about 20 km. The radar head is attached near the electro-optic sensors at the top of the mast mentioned above. Radar such as this is likely to remain desirable for teleoperated ground surveillance, and for use in early warning. Target tracking would likely be a more local activity for the autonomous UGV system, with range requirements of an order of magnitude less, out to say 2000 m. It is also noted that remote or even autonomous deployment of equipment favours simpler mechanisms, due primarily to the difficulty in precise equipment positioning in close quarters. Fully shielded reflectors or perhaps future phased arrays would be encouraged.
Acoustic For outdoor surveillance, acoustic sensors have been used for detection of aircraft (helicopters), and for seismic vibration from moving ground vehicles. For silent surveillance, or more specifically early warning, such sensors should be easily implemented in a UGV Recce system. Low frequency acoustic sensors would not be particularly sensitive to orientation on the UGV, and seismic units could use a fairly simple actuator for ground penetration and coupling. Multiple remote acoustic sensors might also be ‘laid’ and left behind using the teleoperated UGV Recce system.
Landmine Detection The Improved Landmine Detection Project (ILDP) involves using a teleoperated vehicle to bring various sensors to bear on an area of road that is suspected of being mined . In addition to a visible-light camera for general inspection, the remote sensor suite includes a thermal imager to examine heat flow disturbances at the road surface, a road-width metal detector array, a ground penetrating radar array, and a thermal neutron activation detector that can be placed over a suspect location to examine subsurface nitrogen content. This system may resemble near-future UGV Recce systems when they are teleoperated. Efforts are now being made to make the data fusion and mine detection process more automated. However, aside from visible-light and thermal imaging, the other landmine detection technologies are not currently well suited to off-road applications, autonomous deployment or covert missions. The large number of buried landmines (estimated > 100 million) may however encourage sufficient near-term developments to make these technologies more fieldable.
Chemical and Elemental The LAV Recce system  is equipped with a chemical sensor and crew alerting system. A more proactive detection device is used for dismounted point detection . This would seem to be an excellent type of detection and warning capability for an unmanned reconnaissance system. In fact, along similar lines, the teleoperated ILDP rover mentioned above was considered for fitting of a Trace Element Detector (TED). This device, although currently somewhat cumbersome, uses a mass spectrometer to analyze airborne chemical components released form air and soil samples. Teleoperated manipulation of the sampling head could however allow point investigation of dangerous sites.
Radiation The LAV Recce system is equipped with a radiation detector , and alerting system similar to the chemical sensor above. Vehicle-mounted and tele-manipulated sensor heads could be used in parallel with the above chemical sensing system for area and point investigations, respectively.
Nuclear Event Battlefield systems, including manned reconnaissance vehicles, use nuclear event detectors to trigger system protection measures, in case of a nuclear detonation. Radiation hardening and careful attention to electromagnetic pulse shielding might be desirable for unmanned reconnaissance, considering that UGV’s are perhaps the most desirable vehicle to send into an escalating conflict.
Biological Remote air and soil sampling for dangerous biological agents is now desired in some conflicts. Operation Desert Storm required rapid deployment of portable field analysis laboratories. New thin-film membrane and other bio-electrical technologies may allow for portable detectors to be mounted to UGV’s for autonomous early warning. Additionally, teleoperated sampling and analysis might be performed using more elaborate equipment if the need arises.
Navigation and Sighting In addition to the more common surveillance sensors, UGV Recce will require a number of other common sensors for navigation and sighting operations. Navigation will likely use a magnetic compass, GPS, and possibly a combination of odometry and inertial sensors for accurate short term navigation estimates for motion control. Surveillance and sighting activities would certainly make use of GPS to measure position and elevation of the UGV. Sight angle could then be measured using inclinometers, as well as a magnetic compass for approximate azimuth. If more accurate surveying is required, other vehicle-transportable geodetic equipment is available.
4.2 Sensor Deployment
Sensor deployment for reconnaissance or surveillance involves the bringing of a sensor head to bear on the environment in which it is to sense. For UGV Recce, many chassis-mounted sensors might remain deployed at all times, while mast or manipulator-mounted equipment would likely require some amount of extension to properly expose and orient the sensor head. Additionally, some sensors might be particularly delicate or suffer degradation from dirt or unnecessary exposure, and some may require vehicle isolation and active stabilization to perform properly. The following notes consider issues of chassis-mounting sensors, extending them to some distance from the vehicle, and orienting or pointing them.
Chassis Mounting Considering the desire for mobility, stealth and reliability, it would seem that the ideal location for the mounting of reconnaissance equipment would be the inside of the UGV hull. This is of course not suggested to be practical for many surveillance activities, but it would seem to be a good starting point for deployment considerations. Sensors that could in fact be mounted internally include those for UGV engine or other system status monitoring, inertial navigation, vehicle attitude (inclinometers) and even barometric pressure.
The next best mounting location, for the same reasons as stated above, would seem to be on the outside surface of the hull. Indeed this is a reasonable location for many practical sensors, especially those that are omnidirectional in nature, or that may be left in some fixed alignment. Vehicle motion control sensors for proximity or contact could certainly be fixed to the chassis. Mobile robotic systems typically use peripherally mounted arrays of acoustic and infrared transducers, and use appropriate signal processing for directivity. Considering an early warning function of the UGV Recce system, it is noted that chemical, biological and perhaps radiation sensors might not require anything more than hard-mounting with wide-angle external exposure.
Extension The need for extension of a sensor, from the UGV vehicle to the working environment, would seem to be about the least desirable requirement to have to meet, considering the desire for mobility, stealth and reliability, as noted earlier. Nevertheless, most line-of-sight (LOS) sensors will require at least some elevation above the vehicle for reasonable performance, and any point detection activities will likely require some kind of manipulation of a sensor head.
The LAV Recce system  does in fact use turret-mounted LOS sensors, including intensified-visible and thermal imagers for driving, fire-control sighting and surveillance. However, the primary surveillance sensor suite is mounted to the top of a free-standing telescoping mast. This power-driven mast extends from the rear quarters of the vehicle, to elevate a radar unit, imagers and a laser range finder, to a height of about 10 m above the ground. Such deployment generally restricts the vehicle to stationary operation.
Stationary surveillance is intended not only to benefit sensor deployment and performance, but to facilitate covert operations. In fact, it is highly desirable for the reconnaissance vehicle to conceal or camouflage itself within surrounding trees or other close terrain features. Extension and retraction of a mast or other elevating device will likely be a significant reliability problem for UGV Recce systems. Teleoperated deployment in cluttered areas may suffer from lack of feedback during extension, and automated systems will certainly have to be robust, algorithmically as well as mechanically.
In addition to typical surveillance activities, reconnaissance may require site investigations and point detection or sampling for chemical, biological or radiological agents, or to move or manipulate the surroundings during the course of investigation. Fully automated manipulation in the near future would likely be restricted to simple activities, such as driving a coring tube into the soil, or lowering a sensor head close to the ground at some distance from the vehicle. More sophisticated manipulators might however be practical for teleoperated activities.
Pointing and Stabilization The difference between pointing and stabilization control is really just a matter of degree of the same operation. Pointing control usually refers to rather macroscopic placement and orientation of a sensor head, while stabilization refers to keeping the sensor accurately positioned and directed on target using some form of active motion isolation. To facilitate rapid large-scale moves, the pointing device, or director, many not be practically able to achieve the precision in position or velocity control that the sensor requires. A second stage, the stabilizer, may be used to perform small angular displacement correction, and to dampen the system to reduce angular velocities.
For line-of-sight systems, the most common device for pointing control is the pan-tilt platform, which facilitates yaw and pitch (azimuth and elevation) motion. For imagers, the direction and stabilization of roll motion is usually unnecessary, since the base platform can be designed to be acceptably horizontal, image rotation is not usually required, and image sensors do not magnify roll disturbances. The LAV Recce  and other vehicular reconnaissance systems have imagers and laser range finders mounted to pan-tilt platforms for directing their gaze and to track targets. UGV Recce systems will also require directors for both teleoperated and fully autonomous detection and surveillance. Many current pan-tilt devices and related remote operation technologies should be adaptable to UGV activities.
Further active stabilization of systems might be performed by appropriate sensing and feed-back control of the director. If the director characteristics are insufficient to allow for such control, it may be necessary to install an additional stabilizer interface between the base and the director, or the director and the sensor head. Such feedback stabilization was not found to be necessary for the LAV Recce system, however many other systems do in fact require such additional complexities. Noting that UGV Recce systems are likely to be less massive and more structurally adaptable (flexible), it would seem prudent to expect to need to combine some active stabilization with at least all high performance line-of-sight sensors. Operators should have (and may expect) performance similar to established manned systems during teleoperation. Also, near-future autonomous detection and tracking algorithms should not be expected to handle image loss or distortion (ex. image blurring) better than human operators.
4.3 Command, Control and Communication
Autonomy for unmanned vehicles and surveillance, outlined earlier above, would seem to be encouraged not just by the fact of the expected UGV isolation, but by the inadequacies of still developing teleoperation technologies, limited communication bandwidth, and perhaps even manpower constraints. On the other hand, current technologies of autonomous systems are also far too inadequate to allow one to simply instruct the UGV to head out into the field and perform reconnaissance. Taking a lead from decades of systems design for planetary exploration, a blend of the fully independent with directly controlled (teleoperated) machine will be necessary.
Command The commander of an UGV Recce unit will of course not be physically at the surveillance site. While lack of on-site command presence may seem to be a drawback initially, if one can assume that teleoperation and telepresence capabilities for one UGV Recce unit are sufficient, then in fact a single commander might access multiple UGV’s. This not only reduces manpower requirements, but should facilitate more rapid wide-area data acquisition with greater coordination.
It was noted earlier that autonomous capabilities in vehicular and surveillance activities will be limited in near-future systems. In addition, it is still an open question as to how much autonomy would in fact be found acceptable to a commander. Perhaps it should be expected that for single UGV applications, the commander will want fairly direct knowledge and control (via his operator) of the vehicle movement and the surveillance activities, and in a multiple UGV scenario, he would be willing (and required) to give autonomy to many operations while concentrating on the activities of one UGV in a critical situation.
Control While the UGV Recce system commander may be free to work with more than one UGV at a time, it would seem that multiple UGV’s would require multiple operators. At the lowest levels of system autonomy, perhaps two operators per UGV system would be sufficient; one for UGV operations and one for surveillance activities. Eventual elevation in autonomy levels for the vehicle and/or surveillance activities would seem to naturally facilitate reduction to one operator per UGV.
In practical terms, one could therefore see an initial UGV Recce system requiring two operator consoles (vehicle and surveillance), each with a high resolution display, keypads, joysticks, and perhaps a suite of mission recorders (VCR and data recorder). The dual console system would also seem to offer a couple other benefits. Firstly, single operator loads would free one ‘seat’ for observers, namely a commander. Secondly, as autonomy allows one operator to run one UGV Recce system, the second operator then has the equipment to control a second UGV. This last point has yet another interesting systems effect. Assuming sufficient network or other communications are in place, it may be possible to pass UGV’s from one station to another, for logistics reasons or perhaps when a control station must go off-line.
Communication Two paradigms for tele-command of UGV Recce systems are direct transmission, and network messaging. The first appears rather straightforward (traditional), while the second touches on ideas of wide-area networks and the “digitized battlefield”. Communications schemes that tend toward this second idea have some interesting possibilities. First of all, one would expect that all future reconnaissance or even other military communications systems will incorporate encryption of information that is transmitted. With this data security mechanism in place, compatibility with existing commercial and other available communication links can be pursued. This would include virtually all allied satellite and land links, for which the UGV would have to have appropriate transceivers on board.
A few other general concepts, introduced earlier, seem to work well with this networking idea. Increased UGV autonomy allows for a minimum of command and control transmission, and sensor fusion at the UGV would help to keep the returned surveillance data-rate to a minimum. While an UGV can move out of range of some given radio transmitter, low data-rate messages can be passed via network and/or other limited bandwidth systems, regardless of physical location of an UGV or its operators. This brings out the idea that UGV’s could pass from one operational area to another, while remaining under the same operator control, or being transferred to another control station as suggested earlier.
Finally, using a fairly ‘open’ type of secure communication facilitates allied unit reception of surveillance data, without needing full involvement in the UGV Recce system, or having to wait for data processing. Command/control centres may not process information with attention to items which other locations or allied forces have interest. A very similar idea is proposed for air-borne RPV reconnaissance in , using a specialized remote access device. In any case, such a system is then perhaps more of an information resource. Combined with portable computing and sundry digitized battlefield equipment, isolated sections/units/soldiers might be able to utilize surveillance information more directly when necessary.
This paper has discussed the concept of Unmanned Ground Vehicle Reconnaissance (UGV Recce). Beyond line-of-sight land reconnaissance requires ground vehicle activity in hostile areas, where substantial costs of the required systems are incurred by the need to transport operators. Unmanned aerial vehicles, already in use for land reconnaissance, would seem to have limitations for extended covert surveillance. It is suggested that teleoperated UGV’s are appropriate for land surveillance, and that increasing levels of autonomy may ease the operational burden, as well as add new capabilities.
Mobility for UGV Recce is seen to be dictated more by say maneuverability than by acceleration or speed. It is noted that the weight and primary power for UGV Recce vehicles may be very much less than that of the current manned LAV Recce systems, and on-board surveillance systems may therefore require much more power relative to the total budget of future systems. Navigation, encompassing motion planning and control, pulls together vehicular elements of the system, and is perhaps the central location or embodiment of the characteristic of autonomy. Initial UGV Recce systems may move under direct (teleoperated) control, however future developments may allow an operator to issue higher-level mission goals, leaving the UGV to handle all motion planning and execution.
In addition to line-of-sight surveillance, other UGV Recce activities may include detection, monitoring and inspection of explosives (landmines, ordnance) and other chemical, radiation and biological hazards. In addition to chassis- and mast-mounted surveillance equipment, various robotic manipulator and teleoperation technologies may then be required. Finally, although commanders and operators will no longer be on-site, UGV Recce in fact offers expanded capabilities. These include coordinated, multi-UGV wide-area reconnaissance, using a single command point, hand-off of UGV’s from one control station to another, and low data-rate network communications for system control and reception of surveillance data, facilitating long range operation and direct information access by allies.
The author would like to thank Vic Aitken, Chris Brosinsky, Robert Chesney and Doug Hanna of Defence Research Establishment Suffield (DRES), Blair Cain of Computing Devices Canada (CDC) Calgary, and Phil Church and Jim Lougheed of CDC Ottawa, for reviewing the draft paper and making suggestions to improve it. The contributions of numerous other CDC colleagues, mostly through casual conversation, are also gratefully acknowledged.
 Capt. Sean Hooper, “Unit Surveillance, Target Acquisition, and Night Observation Project”, Sensors & Surveillance, Seminar Proceedings, Canadian Defence Preparedness Association, Ottawa, 1996.
 Capt. J. Greengrass, “Unmanned Airborne Surveillance and Target Acquisition System”, Sensors & Surveillance, Seminar Proceedings, Canadian Defence Preparedness Association, Ottawa, 1996.
 Maj. Denis David, “LAV-Recce”, Sensors & Surveillance, Seminar Proceedings, Canadian Defence Preparedness Association, Ottawa, 1996.
 P. Church and T. Gajdicar, “Multi-Window Registration Algorithm for Automatic Detection and Tracking of Moving Targets”, Advances in Guidance and Control of Precision Guided Weapons, Guidance and Control Panel 54th Symposium, Advisory Group for Aerospace Research & Development, AGARD Conference Proceedings 524, NATO, 1992.
 Maria Rey, “Multi-Sensor Data Fusion”, Sensors & Surveillance, Seminar Proceedings, Canadian Defence Preparedness Association, Ottawa, 1996.
 C.E. Thorpe, editor, Vision and Navigation: The Carnegie Mellon Navlab, Kluwer Academic, 1990.
 Proceedings of the IEE / IEEE Vehicle Navigation and Information Systems Conference (VNIS), IEEE, Ottawa, 1993. Also see other locations and years.
 R.A. Brooks, “A Robust Layered Control System for a Mobile Robot”, IEEE Journal of Robotics and Automation, RA-2, pp. 14-23, 1986.
 R.A. Brooks, “New Approaches to Robotics”, Science, Volume 253, pp. 1227-1232, 1991.
 E. Burgess, Return to the Red Planet, Columbia University Press, 1990.
 M.P. Golombek, “Mars Pathfinder: Blazing a New Trail”, The Planetary Report, Vol. XVI, No. 1, The Planetary Society, 1996. Also see Internet World Wide Web page: mpfwww.jpl.nasa.gov
 M.H. Raibert, Legged Robots that Balance, MIT Press, 1988.
 J. Bares, “Ambler: An Autonomous Rover for Planetary Exploration”, Computer, Volume 22, Number 6, pp. 18-26, June 1989.
 Internet World Wide Web page: img.arc.nasa.gov/Dante/
 S.M. Song and K.J. Waldron, Machines That Walk: The Adaptive Suspension Vehicle, MIT Press, 1989.
 V.R. Kumar and K.J. Waldron, “A Review of Research On Walking Vehicles”, The Robotics Review 1, MIT Press, 1989.
 D.J. Todd, Walking Machines: An Introduction to Legged Robots, Kogan Page, 1985.
 D.N. Green, Guidance and Control of a Planetary Rover, M.Eng. Thesis, Carleton University, 1993.
 P. Church, B. Cain and V. Aitken, “Data Fusion Techniques for Accurate Landmine Position Estimation”, TEXPO 96, Computing Devices International, Ottawa, 1996.