Hierarchical Model and Evaluation Method for Autonomy Levels of Unmanned Platforms

Transcription

1 Research Joural of Applied Scieces, Egieerig ad Techology 4(11): , 2012 ISSN: Maxwell Scietific Orgaizatio, 2012 Submitted: December 02, 2011 Accepted: Jauary 04, 2012 Published: Jue 01, 2012 Hierarchical Model ad Evaluatio Method for Autoomy Levels of Umaed Platforms Yibo Li ad Xixig Wag School of Automatio, Sheyag Istitute of Aerospace, Sheyag , hia Abstract: I order to evaluate the autoomy levels of Umaed Platforms, A ew hierarchical model for autoomy levels of umaed platforms ad a ew evaluatio method are provided. Firstly, huma iterface, situatioal awareess, evirometal adaptatio ad decisio-makig of a umaed platform are determied as four idicators applied to evaluate the autoomy levels. Secodly, evaluatio method is used to quatify the four idicators. Fially, the average of the four idicators is the value o behalf of the autoomy level. The ew model has bee applied to judge the autoomy level of Global Hawk USA, Red FlagHQ3 UGV ad Sparta Scout USV. The results show that the four idicators are comprehesive ad visualized, ad the evaluatio method is simple. Key words: Autoomy level, evaluatio method, hierarchical idicators, umaed platform INTRODUTION With the developmet of computer ad commuicatio techologies, the complexity ad automatio degree of umaed platforms is icreasig day by day. The highly dyamic, ucertaity ad the complexity of missio ad flight coditios of applicatio eviromet reveal that the autoomy level has become a ew challege faced by umaed platforms (Yag ad Zhag, 2009). At preset, may coutries have struggled for years to research the autoomy levels of umaed platforms. However, the restrictios of the techical level ad some other coditios prove that it is impossible for umaed platforms to achieve the full autoomy without the perso s supervisio. Therefore, it is imperative to establish a hierarchical model for autoomy levels of umaed platform. The correct assessmet of the autoomy level will make umaed platforms complete the task efficietly ad timely, ad i the process of accomplishig the task, the cost is the lowest ad the reliability is the highest. Los Alamos uses mobility, acquisitio, ad protectio to measure the autoomy level of umaed platforms, but the metrics do ot address operatioal characteristics of umaed platforms ad ot show the differece amog differet autoomy levels (lough, 2002); harles Stark Draper utilizes mobility cotrol, task plaig ad situatioal awareess to describe the autoomy level of umaed platforms (leary et al., 2000), however, task plaig is t a prerequisite to the autoomy, it is just the reactio to complicated situatios. Situatioal awareess is oly based o the umber ad iformatio fusio of sesors, which does ot reflect whether the umaed platforms have uderstood what s goig aroud the eviromet or ot. The ALFUS idetifies that the autoomy levels for umaed platforms must be characterized through the followig three perspectives: the missios that the UMS is required to perform or is capable of performig, i the kids of eviromets, ad with the levels of huma iteractio (Huag et al., 2003a,b). They devise a three-axis represetatio for the cocept (Huag et al., 2005a,b). Each axis is elaborated with a set of metrics. But this method is ot directly measure the autoomy levels of umaed platforms, oly by determiig the relatioship betwee exteral ad umaed platform to measure the autoomy, is ot reliable. For istace, a umaed platform is remotely cotrolled for a very complex missio i a very difficult eviromet, but we ca t warrat that its autoomy level is high eough (Huag et al., 2004a,b). Metrics for each axis ad the weight distributio are also ill-fouded. As a result, this paper presets a ew hierarchical model ad evaluatio method for autoomy levels of umaed platforms. METHODOLOGY Hierarchical model: The hierarchical model cosists of two compoets: idicator model ad level model. The Idicator model is elaborated with a set of idicators determiig the autoomy level of umaed platforms. The level model is the short descriptio of each autoomy level. Idicator model: The idicator model icludes four kids of hierarchical idicators: huma iterface, situatioal awareess, evirometal adaptatio ad decisiomakig. orrespodig Author: Yibo Li, School of Automatio, Sheyag Istitute of Aerospace, Sheyag , hi 1488

2 Res. J. Appl. Sci. Eg. Techol., 4(11): , 2012 Huma iterface icludes four sub-idicators as followig: Operator cotrol time: i the process of performig missio, the loger the operator cotrols the platform, the lower the platform s autoomy level is. Accordigly, the poits o behalf of operator cotrol time are much lower. Full poits are 100. For example, whe performig a missio, half of the time is cotrolled by operators, the this sub-idicator ca get 50 poits. If the operators do t eed to participate i the process all the time, the this sub-idicator ca get 100 poits. The poits are determied by people ad ot calculated by mathematical formula. Therefore, the poits are subjective: Operator cotrol umber: the more umber of platforms a operator cotrols, the higher the platform s autoomy level is, ad this sub-idicator will get higher poits. Operator skill level: the operators skill level idirectly reflects the autoomy level for umaed platforms. The divisio criteria for skill level are the operators qualificatios i this paper. Operator workload: this workload maily refers to the focused itesity of a operator durig the cotrollig time. Situatioal awareess icludes four sub-idicators as followig: Attack type: the type of attacker maily icludes aerial (UAV), groud (UGV), uderwater (UUV), surface (USV), etc. The more types a umaed platform idetifiers, the higher poits the subidicator gets. Attack frequecy: it maily relates to the umber of perceivig ad gettig away from attackers i a fixed period of time. Attack risk idex: it relates to the ability of perceivig dagerous degree of the attacker. Reactio time: it relates to the time of perceivig dager of umaed platforms. The shorter the reactio time is, the higher poits the umaed platform gets. Evirometal adaptatio icludes three subidicators as followig: Eviromet type: the more types a umaed platform ca adapt to, the higher poits the subidicator gets. Adaptatio time: it relates to the time of adaptig to differet atural eviromets. The shorter the adaptatio time is, the higher poits the umaed platform gets. Adaptatio scope: it relates to the scope of atural eviromet perceived by a umaed platform. Huma iterface icludes five sub-idicators as followig: Attack type: this sub-idicator is maily used to evaluate the ability of real-time diagosig the health of umaed platforms, especially i the process of performig missios. Fault self-repair: it relates to the ability of umaed platform solvig faults whe the faults occur. Path plaig: it relates to the ability of dyamic path plaig ad re-plaig of the umaed platform whe uexpected states occur i the process of performig missios (he, 2007). Task plaig: it relates to the ability of dyamic task plaig ad re-plaig of the umaed platform whe uexpected states occur i the process of performig tasks. ooperatio: whe a variety of umaed platforms collaborate to complete the commo task, there are may uexpected states i perceptio, performace, commuicatio, etc. Therefore, how to realize the high-level autoomy collaboratio amog the umaed platforms ad complete the task assiged as well efficietly is quite importat. Level model: The geeral developmet tred of umaed platforms is remote cotrol-semi-autoomousfully autoomous (Gao et al., 2007). So the level model is divided ito three levels accordig to the developmet tred, amely, remote cotrol, semi-autoomy ad full autoomy. Each level is described a rage of scores. The full scores are 100 poits. Remote cotrol presets that umaed platforms oly operate accordig to a predetermied program. The software ca oly deliver the perceptio data to the operator for processig ad ot idetify ad uderstad the situatio of the eviromet. The umaed platform does t have access to decisio-makig. The perceptio tools ad computig tools i hardware are primary ad oly perform the most basic fuctios ad o weapos carryig capacity. Semi-autoomy demads that umaed platforms ca realize autoomy i some fuctios. The software ca distiguish betwee static ad dyamic eviromet, idetify weather coditios, uderstad operate eviromet, summarize the curret status of the flight ad the eviromet, perceive the curret health status ad have the basic weapos carryig capacity. The predetermied program ca be chaged partly to meet the curret uexpected coditios; Hardware perceptio tools 1489

3 Res. J. Appl. Sci. Eg. Techol., 4(11): , 2012 ad computig tool have also reached the medium level ad ca support to ru more advaced autoomous software. Full autoomy demads that umaed platforms ca realize autoomy i all of the fuctios. The software ca ot oly distiguish betwee the exteral ad iteral eviromet, but also realize resources reorgaizatio ad the real-time plaig of paths ad tasks i order to make the curret state of a umaed platform achieve optimal. The hardware ca support the complexity, highlevel autoomous program. Evaluatio method: The evaluatio method is used to quatify the four idicators of a umaed platform, ad gets the value o behalf of the autoomy level. Fuzzy comprehesive evaluatio is the mai core i the evaluatio method. There are four kids of idicators, so this paper just takes the quatitative process of uma iterface idicator for example. Factor set U: The factor set of sub-idicators of huma iterface as follows: U 1 = (Operator cotrol time, Operator cotrol umber, Operator skill level, Operator workload) = (u 11,u 12,u 13.u 14 ) Evaluatio set V: V= (remote cotrol, semi-autoomy, full autoomy) = (v 1,v 2,v 3 ) I this study, the levels of autoomy are separately remote cotrol, semi-autoomy, full autoomy. The highest scores of remote cotrol are 20 poits. The highest scores of semi-autoomy are 60 poits. The highest scores of full autoomy are 100 poits. Each level score is writte i vector Q= (20, 60, 100). Weight vector W: This study uses the Aalytic Hierarchy Process (AHP) to calculate the weight of each sub-idicator of uma iterface ostruct judgmet matrix A: Because the subidicators of uma iterface have bee give differet poits, the pairwise compariso method is used to establish the judgmet matrix A 1 of uma iterface accordig to the differet poits. 1 a12 a13 a14 1 a12 1 a23 a24 A1 / 1/ a13 1/ a24 1 a34 1/ a14 1/ a24 1/ a34 1 (1) a ij is the relative importace of the i idicator compared with the j idicator. The value of a ij is geerally a positive iteger 1-9 (called scale) or their iverse. If there is a umaed platform, the scores of operator cotrol time u 11 are subjective give 50 poits; the scores of operator cotrol umber u 12 are 25 poits; the scores of operator skill level u 13 are 35 poits; the scores of operator workload u 14 are 45 poits. Accordig to the value rules of a ij, u 11 compared with u 12 is obviously importat, the a 12 = 5; u 11 compared with u 13 is slightly importat, the a 13 = 3; u 11 compared with u 14 is ot slightly importat eough, the a 14 = 2. alculate the weight W: The steps of calculatig the weight as follows: Normalizig the colum of the judgmet matrix ij A ij ( ) i1 ij Summig the elemets of each row of A ij : W a a j a 2 j j,,..., j1 j1 aij aij aij 1 j1 i1 i1 i1 T (2) After the ormalizatio of w, the weight vector ca be got that. W = (w 1, w 2, w 3,, w ) (3) The weight ca chage with the scores of each subidicator of huma iterface. Membership matrix R: The results that each idicator i the factor set U 1 = (u 1, u 2,, u ) is measured by each elemet i the evaluatio set V= (v 1,v 2,v 3 ) costitute the membership matrix R. r11 r12 r1 r r r R rm1 rm2 rm (4) r ij is the result that the j idicator measured by the i elemet. Each idicator is give a subjective score. Therefore, i the paper, membership fuctios are costructed to calculate the membership that the idicator belogs to every autoomy level accordig to the subjective scores (0#x#100). The differet scores result that the idicator s membership will also be differet. 1490

4 Res. J. Appl. Sci. Eg. Techol., 4(11): , 2012 The membership fuctio that x belogig to remote cotrol: 80 x 0 x 80, 1 ( x) 80 0, x 80 (5) The membership fuctio that x belogig to semiautoomy: 0, 0 x 10, x 90 x 10 2 ( x), 10 x x, 50 x (6) The membership fuctio that x belogig to full autoomy: 0, 0 x 20 x x 20 3( ), x (7) For example, the sub-idicators of uma iterface have bee give the score subjectively. The scores of operator cotrol time u 11 are give 50 poits, the: r 1 1 = 1 (x) = 1 (50) = (8) It shows that whe operator cotrol time u 11 is give 50 poits, the membership that the sub-idicator belogs to remote cotrol is 0.375: r 12 = 2 (x) = 2 (50) = 1 (9) u 11 It shows that whe operator cotrol time is give 50 poits, the membership that the sub-idicator belogs to remote cotrol is 1: r 13 = 3 (x) = 3 (50) = (10) It shows that whe operator cotrol time u 11 is give 50 poits, the membership that the sub-idicator belogs to remote cotrol is As a result, the membership vector of operator cotrol time u 11 is: r 1 = (r 11, r 12, r 13 ) = (0.375, 1, 0.375) (11) Fuzzy decisio B: Usig the membership matrix R 1 ad weight vector W 1 that have calculated to make 1-level fuzzy trasformatio to get the decisio set B 1 (Luo et al., 2010) that the membership vector of uma iterface belogig to evaluatio set V: B = W R = (b 1, b 2, b 3 ) (12) Table 1: Score of each idicator of global hawk Hierarchical idicators Scores (0# x#100) Operator cotrol time u Operator cotrol umber u Operator skill level u Operator workload u The fial scores of uma iterface are: S 1 = B Q = (b 1,b 2,b 3 ) (20,60,100) (13) Similarly, the fial scores of situatioal awareess S 2, evirometal adaptatio S 3 ad decisio-makig S 4 ca be got. EXAMPLE OPERATION The ew model has bee applied to judge the autoomy level of Global Hawk USA, Red FlagHQ3 ad Sparta Scout USV. Take the calculatig process of Global Hawk USA for example. Table 1 is the subjective score of each idicator of Global Hawk. Accordig to the above quatitative process of uma iterface idicator ad the score of each idicator of Global Hawk, the fial quatitative scores of the four idicators of Global Hawk ca be got. The fial score o behalf of autoomy levels of Global Hawk is the average score of uma iterface situatioal awareess evirometal adaptatio ad decisio-makig The scores of the four idicators of Global Hawk as follows: The score of uma iterface idicator is the score of situatioal awareess idicator is the score of evirometal adaptatio idicator is the score of decisio-makig idicator is The fial score o behalf of autoomy level of Global Hawk is I the Uited States UAV road map, the autoomy level of Global Hawk is 2 to 3 (Umaed Aircraft System Roadmap, ). This autoomy level reveals that Global Hawk has achieved adaptig to failure ad flight coditios. But the (Autoomy otrol Level) AL does t provide that umaed platforms how to achieve the autoomy level ad how to quatify the fuzzy idicators. The ew model ca just solve the defects of AL. As a result, the ew model is more coveiet i practice. Similarly, the scores of the four idicators of Red FlagHQ3 UGV as follows: The score of uma iterface idicator is The score of situatioal awareess idicator is The score of evirometal adaptatio idicator is

5 Res. J. Appl. Sci. Eg. Techol., 4(11): , Sparta scout Global Hawk Evirometal adaptatio Red flaghq 3 Situatioal awaeress Decisio-makig 100 Huma iterface Fig. 1: The autoomy level pyramid chart Table 2: Level model for umaed platforms Autoomy level Autoomy score Remote cotrol [0,20) Semi-autoomy [20,60] Full autoomy (60,100] The score of decisio-makig idicator is The fial score o behalf of autoomy level of Red FlagHQ3 UGV is The scores of the four idicators of Sparta Scout USV as follows: The score of uma iterface idicator is 5.12 The score of situatioal awareess idicator is The score of evirometal adaptatio idicator is The score of decisio-makig idicator is 9.20 The fial score o behalf of autoomy level of Sparta Scout USV is Accordig to the Table 2, the autoomy level of Global Hawk is semi-autoomy; the autoomy level of Red FlagHQ3 UGV is full autoomy; the autoomy level of Sparta Scout USV is remote cotrol. I order to kow the four idicators how to play a impact o the autoomy level of umaed platforms, the quatitative scores of the four idicators have bee marked i the pyramid chart. It is show i Fig. 1. The top of the pyramid is 0 score, ad the bottom of the pyramid is 100 scores. The first shaded layer is the four idicators score of Sparta Scout USV. The secod shaded layer is the score of Global Hawk USA. The third shaded layer is the score of Red FlagHQ3 UGV. ONLUSION A ew hierarchical model for autoomy levels of umaed platforms ad evaluatio method are costructed i the study. The idicator model i the hierarchical model ot oly has the direct idicator, like as decisio-makig, but also has idirect idicator, like as huma iterface. The ew mode is ot simply usig sesors ad other devices to idetify the evirometal coditios, but uderstad the curret eviromet. Exteral factors ca t play a large impact o the evaluatio of autoomy level. The fuzzy comprehesive evaluatio ad AHP are applied to quatify the four idicators ad get the scores o behalf of the four idicators. The average of the four quatified scores is the fial scores o behalf of the autoomy level. Accordig to the level model, the autoomy level of a umaed platform ca be decided. The ew model ad method are maily applied to evaluate the sigle umaed platform ad ot suitable for umaed platform group. The autoomy level model is ot very complicated, ad the user ca chage it accordig to the practical applicatio. The score of each idicator is decided by the people, ad it is subjective ad ot fixed score. REFERENES he, H., Summary of UAV Autoomous otrol ad Autoomous Ladig otrol system Xi a: School of Aeroautics, Northwester Poly Techical Uiversity. leary, M., M. Abramso, M.B. dams ad S. Kolitz, Metrics for Embedded ollaborative Itelliget Systems. harles Stark Draper Laboratory, MA, USA. pp: lough, B.T., Metrics schmetrics! How the heck do you determie a UAV s autoomy ayway. Proceedigs of the performace metrics for itelliget systems workshop, pp: 1-7. Gao, J., Q. Zou ad S. he, Study o the cocept of autoomy for UAV. Elect. Opti. otr., 14:

6 Res. J. Appl. Sci. Eg. Techol., 4(11): , 2012 Huag, H., M. Elea ad A. James, 2003a. Toward a geeric model for autoomy levels for umaed systems (ALFUS). Proceedigs of the performace metrics for itelliget systems workshop, pp: Huag, H., A. James ad M. Elea, 2003b. Autoomy level specificatio for itelliget autoomous vehicles: Iterim progress report. Proceedigs of the Performace Metrics for Itelliget Systems Workshop, pp: Huag, H.M., A. James, M. Elea, R. Wade ad W. Eglish, 2004a. Specifyig autoomy levels for umaed systems: Iterim report. Proceedigs of SPIE Defese ad Security Symposium, pp: Huag, H.M., E. Messia, R. Eglish, R. Wade, J. Albus ad B. Navak, 2004b. Autoomy measures for robots. Proceedigs of the 2004 ASME Iteratioal Mechaical Egieerig ogress ad Expositio, Aaheim, aliforia, pp: Huag, H.M., P. Kerry, A. James, B. Novak ad M. Elea, 2005a. A framework for autoomy levels for umaed systems (ALFUS). Proceedigs of the AUVSI s Umaed Systems North America, pp: Huag, H.M., P. Kerry, A. James, ad M. Elea, 2005b. Autoomy levels for umaed systems (ALFUS) framework: A update. Proceedigs of SPIE Defese ad Security Symposium, pp: Yag, Z. ad R. Zhag, Fuzzy evaluated method for the autoomy levels of umaed systems. J. hiese omp. Syst., 10: Umaed Aircraft System Roadmap, Office of the Secretary of Defese. Washigto D. Luo, Z., H. he ad N. Zheg, Veture ivestmet project appraisal based o multi-factor fuzzy sythetic evaluatio model. J. xi a Uiv. Sci. Techol., 30: