AUVs: In Space, Air,Water, and on the GroundBy David A. Schoenwald

Autonomous unmanned vehicles (AUVs) have gener-ated much interest in recent years due to their greatpromise for performing repetitive, dangerous, or in-

formation-gathering tasks in hazardous or remote environ-ments. The diversity of environments in which these vehiclesmust operate (space, air, water, ground, and combinationsthereof) results in a wide variety of vehicle types. In addition,the requirement for autonomy from direct human interactionplaces heavy emphasis on reliable control strategies. Thisspecial issue presents four articles that focus on controlstrategies for each of the four types of terrain.Although these articles represent just atiny sampling of research in thefield of AUVs, they should givethe reader an understandingof some of the approaches be-ing taken in the design, model-ing, and control of thesevehicles, as well as the manychallenges facing designers.Among these challenges are communi-cations, sensors, materials, locomotion,end-effectors, cooperative control and machine-level con-trol, as well as computational horsepower. Each article ad-dresses several or all of these issues, as well as possibleapproaches. Other special issues (e.g., [2]-[4]) have also ad-dressed these challenges for specific types of AUVs.

Need for AUVsThe primary reason for considering the use of AUVs is theirability to gather information, manipulate physical objects, orengage some kind of equipment in remote or hazardous loca-tions. The types of applications envisioned for AUVs includeenvironmental remediation, detonation/defusing of live am-

munition, navigation within an area to gather data, militaryreconnaissance, transportation of goods, performance of re-petitive and dangerous tasks, and the like. The developmentof AUVs was originally driven by the need for remediation ofhazardous waste sites in which human intervention wascostly and dangerous. Although that is still a driving force,there is also the U.S. military’s need for intelligence gatheringand operational support in the face of reduced manpower. In-deed, the shortage of personnel and the high cost of laborhave become major factors in the need for autonomous vehi-

cles. Transportation—automated highways,in particular—has also been a significant

area of research in AUVs; this specialsection does not include such

work, however, since it has beendealt with before (see the Octo-ber and December 1996 issuesof CSM). The articles here focus

more on unmanned vehicles. Insome environments, such as space

and certain underwater locations, itmay be nearly impossible to bring humans

into the area; thus, the need to do any exploration or retrievaldepends crucially on AUVs. The first article, by Carignan andAkin, describes such a situation in space.

Technical Challengesin AUVsBy their nature, autonomous vehicles need to understandenough about their surroundings so that they can functionwith minimal or no input from humans. This implies sensorsare needed that are capable of “seeing” terrain as well asidentifying obstacles, other vehicles, and any goals or tar-gets they are approaching. To control these vehicles, useful

December 2000 IEEE Control Systems Magazine 15

Guest EDITORIAL

The author (daschoe@sandia.gov) is with Sandia National Laboratories, P.O. Box 5800, MS 1004, Albuquerque, NM 87185, U.S.A.

0272-1708/00/$10.00©2000IEEE

models of their dynamics are needed in a variety of mathe-matical forms, depending on the type of control desired.In addition, simulation of these models is necessary forvehicles that are expensive to operate. This approach canalso provide a proof of concept to potential customers ofthe vehicle.

The issue of locomotion or propulsion is a complex onethat relates to the vehicle’s ability to handle the terrain it isintended to negotiate. For instance, a stair-climbing robotcould use a variety of crawling, walking, hopping, or hover-ing maneuvers, but this may not be the best means for goingthrough a forest or across a desert. Further, any tasks thesevehicles are required to perform may require additional ma-nipulation. Consider the situation of using an AUV to investi-gate a car that is believed to be wired with explosives. First,some mobility would be needed to get to the vehicle. Ofcourse, sensors would be needed to detect the presence ofexplosives. The act of opening the door (or some othermeans of entry) would require some articulated manipula-tion of an appendage. Finally, any means to investigate ordetonate the explosives would require further manipula-tion. Thus, the vehicle may need a variety of locomotion/manipulation capabilities. The design of AUVs entails manyother issues, including materials, battery power needs, andpayload capabilities. Further discussion of these issues canbe found in [2]-[4].

The control of autonomous vehicles will depend heavilyon the sensors, locomotion/propulsion, onboard process-ing power, and dynamic models available. The type of ter-rain (air, water, space, ground) the vehicle must traversepresents its own set of problems. Table 1 lists the four pri-mary environments, along with a few of the technical chal-lenges and potential pitfalls facing AUV designers. For themost part, relative terms are used to describe the AUV at-tributes so that Table 1 can be used for comparison pur-poses. Although space and underwater vehicles may havemore favorable buoyancy and gravity environments, thesecan be particularly harsh environments in which to operatea vehicle. Further, communication from a vehicle to a hostor from vehicle to vehicle is especially difficult, particularlyfor underwater vehicles. This may require the vehicle to op-erate in a “batch” mode. Aerial (and to a lesser extentground based) vehicles must deal with gravity forces, butthe physics of their interactions with the environment are

much better understood. Thus, detailed models can be de-rived for these situations.

Much prior work on AUVs has been centered on ground-based vehicles. These have included both wheeled andwalking machines (see the article by Moore and Flann inthis issue) with dimensions of just a few inches on up tothose the size of large trucks. Much work has been done onunderwater vehicles as well. The dynamics associatedwith underwater motion experienced by these vehicles areconsiderably more complex (and nonlinear) compared tothose their ground-based brethren experience. The model-ing and control of these vehicles present many challengesand have led to interest in areas such as nonlinear systemidentification and neural networks [6]. Currently, work isprogressing steadily in enhancing the sensory capabilitiesof AUVs. The entire field of autonomous navigation hasseen a great deal of activity (see [2]). Machine vision hasbeen a particularly promising area, one that it is hoped willgive AUVs the ability to recognize obstacles, targets, andpotential paths to traverse.

The concept of multiple AUVs working together toward acommon goal has begun to generate interest in the autono-mous vehicle community. The goals for these efforts in-clude the ability to search large areas with little or nohuman intervention and to gather material, information,and the like from remote or hazardous locations. MultipleAUVs can be more efficient than a single vehicle. By spread-ing out across the terrain, they can search a large area quiterapidly. Thus, there is a need for intervehicle communica-tion that lets each vehicle know the overall status of the op-eration and whether the specific searching criteria havechanged. Each AUV will still need onboard sensors and con-trols to navigate its assigned search space, but there is alsoan interaction component to their control that begs theneed for concepts from decentralized control theory.

Recent work has taken many different approaches. Thestrategies employed are based on diverse fields such as artifi-cial intelligence, game theory, biology, distributed control,and genetic algorithms. Some of these investigations haveused large-scale system control theory (see, for example, [1]).At Sandia National Labs (http://www.sandia.gov/isrc/sdr_program.html and [5]), models and simulations are being de-veloped that lead directly to the capability to simulate and un-derstand the introduction and behavior of swarms of

16 IEEE Control Systems Magazine December 2000

Table 1. Characteristics of AUVs.

Attribute/Media Ground Based Aerial Underwater Space

Vehicle size 1 cm–10 m 10 cm–10 m 10 cm–100 m 1 m–10 m

Dynamic model complexity Simple to complex Standard Complex Standard

Communication options Numerous Numerous Very restrictive Limited

Environment Easy to difficult Moderate Difficult Severe

Degree of autonomy Full, tethered,teleoperated

Full, teleoperated Full, tethered Full, attached tolarger vehicle

semiautonomous robotic agents in urban and military envi-ronments. For very large numbers (tens of thousands) of au-tonomous agents, physics-based modeling of swarms ofautonomous robots can be used. These approaches employconcepts from statistical mechanics, molecular dynamics,and plasma physics. The advantages of large-scale systemsperspective include leveraging a large body of work on stabil-ity, fluctuation spectra, equilibrium, and efficient computationof the dynamics of potentially large ensembles of interactingobjects. Due to the nature of these applications, the controltechniques must be distributed, and they must not rely onhigh-bandwidth communication between agents. At the sametime, a single host must be able to easily direct theselarge-scale systems. Finally, the control techniques must beprovably convergent so as not to cause undue harm to nearbyhumans and structures.

As an example of a cooperative task, Fig. 1 illustrates agroup of such vehicles that have “surrounded” a target togather information or perhaps to prevent entry/escape.The vehicles have the ability to communicate in a tokenring in which each vehicle has a specified time slot duringwhich to talk. If a vehicle becomes incapacitated, the ringis reduced (after waiting a predetermined number of cy-cles) by one, and the remaining vehicles continue commu-nication and control. The various types of coordinatedbehavior include dispersion, clustering, orbiting, and fol-lowing; that is, the vehicles can spread out to cover aspace or region, group together to investigate a particulartarget, circle around a region or building, and move for-ward in a fairly straight-line progression, respectively.The vehicles may change these goals as one or more ofthem discover new information and communicate it toeach other and/or to a host. Fig. 2 shows a group of similarvehicles following each other in a linear progression. Thisbehavior can be the start of a more complex maneuverthat can include dispersion, orbiting, and clustering withrules that determine when each vehicle should switchfrom one mode to another.

Fig. 3 provides a snapshot of a simulation in which the ve-hicles are spreading out over a terrain to achieve maximumsearch coverage of the landscape while maintaining RF com-munication with each other and the host. This example illus-trates the importance of modeling and simulation with alarge number of AUVs due to the difficulty and expense ofhardware tests. The simulation can show how specific con-trol and communication strategies can be implemented, aswell as possible stability/convergence issues. Each vehiclehas its own control strategy that may consist of a hierarchyof behaviors (maintain RF, triangulate location, detect andavoid obstacles, follow a prescribed path, etc.). In addition,the overall group has a control strategy that might be pre-programmed into each vehicle or communicated via a host.The simulation provides graphical results of the group’smovements as well as data that can be analyzed. It is impor-tant that the simulation include enough vehicle dynamics tobe “physics based” while not being too complex to preventreal-time motion. The ability to quickly change parametersand re-simulate is also necessary.

Special Section OverviewThe four articles that follow address various issues associ-ated with autonomous vehicles. Each article investigates adifferent medium for these vehicles: space, air, water, andground. Two of the four articles examine the multiple-vehicleproblem (air and water); the others focus on a single-vehiclescenario. All articles present experimental and/or simulationresults of the control strategies employed, which includeadaptive control, decentralized control, and optimal (LQR)control. The dynamics of the vehicles are also derived fromphysics-based arguments in each article. Finally, all four arti-cles discuss the merits and deficiencies of their chosen strat-egies and propose numerous ideas for future study.

The first article, by Carignan and Akin, addresses the is-sue of stabilizing space-based robotic vehicles. These ro-bots may be attached to a larger spacecraft or be capableof some free mobility. The article describes a detailed

December 2000 IEEE Control Systems Magazine 17

Figure 1. A trio of RATLERs (Robotic All Terrain LunarExploration Rovers) at Sandia National Labs’ Intelligent Systems &Robotics Center has surrounded a target.

Figure 2. A group of RATLERs searching a terrain via afollowing maneuver.

model of free-flying robots, along with arm dynamics inspace, as well as real-time simulations of the model. Theauthors also discuss the issue of attitude stabilization ofthe spacecraft using arm operations. Parallels are drawnbetween space and underwater robots using examples ofboth. The authors encourage the two robotic communitiesto partner in future endeavors.

The article by Giulietti, Pollini, and Innocenti is concernedwith a multitude of unmanned aerial vehicles (UAVs) flying ina formation. The authors describe various formation typesfor aerial operations and seek to optimize these formationswith respect to communications between the UAV nodes.They also study the consequences and remediation of thefailure of a UAV or communication between UAVs. The articledevelops a UAV dynamic model, then details formation typesand control and communication issues under failures anddropouts, and finally presents some simulation results. A se-ries of figures illustrates the various strategies described.

The third article, by Stilwell and Bishop, focuses on a pla-toon of underwater vehicles. Rather than explicitly specifythe position of each vehicle in the platoon, the authors con-trol the global functions of the platoon, such as its centerand the distribution of vehicles around the center. Decen-tralized control laws are used to achieve this platoon con-trol with little need for communication among the vehicles.This also reduces the sensory requirements of each vehicle.This strategy allows scalability to a larger number of vehi-cles in the platoon with few additional modifications. Simu-lation results for platoons are presented.

The AUV platform considered in the last article, byMoore and Flann, is a six-wheeled ground-based mobile ro-bot. The vehicle is omnidirectional, with each wheel con-trolled independently, and thus it can completely control itsorientation and motion in a plane. The authors describe thevehicle hardware, as well as its electronic architecture—in-deed, the article constitutes a good design study of these as-pects. The mission and path planning for the vehicle isdescribed via a task-decomposition approach. The trackingcontrol is achieved with the use of feedback linearization.Both simulation and experimental results of the planningand control strategies are presented. Numerous figures de-pict the vehicle hardware in various levels of detail.

ConclusionsThis special section is intended to acquaint the reader withcurrent research in the field of AUVs. Since it is such a largeand diverse field, these four articles provide only a brief in-troduction. Indeed, each medium (space, air, water,ground) presents its own set of issues and challenges for ve-hicles that must traverse that terrain. Further, most of theresearch in these four areas has progressed separately. Thehope is that inclusion of an article from each area will en-courage researchers to “borrow” ideas from other types ofautonomous vehicles for use in their own areas. The appli-cations for these vehicles are limited only by one’s imagina-tion. The design, modeling, and control of such vehiclesinvolves virtually every facet of scientific and engineeringendeavor. Thus, this should be a fertile field of research fora long time to come.

AcknowledgmentThe guest editor wishes to extend thanks to all the authorsfor their contributions, to the referees for their constructivecomments, and to the editor-in-chief and his staff for theirencouragement and assistance on this issue.

References[1] D.D. Siljak, Decentralized Control of Complex Systems. San Diego, CA: Aca-demic, 1991.[2] B. Bhanu, O. Faugeras, B. Sridhar, and C.E. Thorpe, “Introduction to thespecial section on perception-based real-world navigation,” IEEE Trans. Ro-botics Automat., vol. 10, pp. 725-727, Dec. 1994.[3] A.K. Bejczy, S.T. Venkataraman, and D. Akin, “Introduction to the specialissue on space robotics,” IEEE Trans. Robotics Automat., vol. 9, pp. 521-523,Oct. 1993.[4] J. Yuh and T. Ura, “Guest editors’ introduction, special issue on autono -mous underwater robots,” Autonomous Robots, pp. 75-77, June/July 1996.[5] J.T. Feddema, C. Lewis, P. Klarer, and R. Caprihan, “Cooperative roboticsentry vehicles,” in Proc. SPIE, vol. 3839, Boston, MA, Sept. 1999, pp. 44-54.[6] J. Farrell, T. Berger, and B. Appleby, “Using learning techniques to accom -modate unanticipated faults,” IEEE Contr. Syst. Mag., vol. 13, pp. 40-49, June1993.

18 IEEE Control Systems Magazine December 2000

Figure 3. A computer simulation of a group of RATLERsdispersing across a hilly terrain.