Terrain Navigation Through Knowledge-based Route Planning

The advent of advanced computer a r c h i t e c t u r e s fo r p a r a l l e l and symbolic processing has evolved to the po in t where the technology c u r r e n t l y e x i s t s fo r the development of p ro to type autonomous v e h i c l e s . Contro l o f such devices w i l l r equ i r e communication between knowledgebased subsystems in charge of the v i s i o n , p l ann ing , and c o n f l i c t r e s o l u t i o n aspects necessary to make autonomous veh ic les f u n c t i o n a l In a rea l wor ld envi ronment . This paper descr ibes a h e u r i s t i c rou te p lanning system capable of forming the p lanning foundat ion of an autonomous ground v e h i c l e . An e f f e c t i v e route planner must address two leve ls of p l ann ing . At the high l e v e l , the p lanning a lgo r i t hm usua l l y e x p l o i t s some form of map data to generate a g lobal route which avoids mountains, v a l l e y s , and canyon a reas . Depending upon the map r e s o l u t i o n , I n d i v i d u a l p i xe l path po in t s at t h i s leve l vary from a rep resen ta t i on area of 100 by 100 meters to 12.5 by 12.5 meters . At the low l e v e l , the p lanning a lgo r i t hm must deal w i th local p lanning ins ide these i n d i v i d u a l path po in t s so t h a t obs tac les such as t r e e s , r ocks , and holes may be avo ided. A consensus e x i s t s t ha t g loba l i n fo rmat ion in the form of a d i g i t i z e d t e r r a i n map Is essen t i a l in the opera t ion of an autonomous v e h i c l e , at leas t at the g lobal leve l of rou te p lanning [ 1 4 ] . The t e r r a i n map requ i red must be capable of represen t ing huge amounts of m u l t i dimensional data In a format e a s i l y e x p l o i t e d by the rou te p lanner . In order to make it s u i t a b l e as a basis fo r r e a l t i m e dec is ion making dur ing veh i c l e o p e r a t i o n , the data should a lso be preprocessed In to a more compact and manageable fo rmat . Two I n t e r e s t i n g approaches to t h i s problem are the ru le-based c r e a t i o n of a "composi te map" [3] and the t r a n s f o r m a t i o n of a polygonal obs tac le map In to a r e l a t i o n a l graph database [ 1 ] . I t Is to the second concept t ha t our c u r r e n t approach to preprocessing map data Is more c l ose l y a I igned . This paper descr ibes a dynamic rou te p lann ing and execut ion system c a l l e d TREK (Tech Route Execut ion Kale do scope) . TREK cons i s t s o f four d i s t i n c t processing phases: (1) scene r e c o g n i t i o n , (2) route g e n e r a t i o n , (3) scene match ing, (4) and knowledge-based v a l i d a t i o n . Each of these processes Is discussed in the f o l l o w i n g sect Ions . The r e c o g n i t i o n of ob jec ts in an image is cen t ra l to the concept of local rou te p l a n n i n g . Given tha t a g loba l rou te can be generated through a large area of t e r r a i n , scene r e c o g n i t i o n must be capable of i n t e r p r e t i n g the ob jec ts conta ined w i t h i n the cu r ren t f I e I d o f v i e w so t ha t they may be avoided dur ing the t r a v e r s a l of the actual t e r r a i n . Without a c a p a b i l i t y to determine what is in i t s local env i ronment , low level route p lanning would not be able to avoid t h r e a t e n i n g ob jec ts or ob jec ts o b s t r u c t i n g the v e h i c l e ' s pa th . Scene c l a s s i f i c a t i o n in the TREK system e x p l o i t s a h i e r a r c h i c a l c l a s s i f i c a t i o n t r e e c o n s i s t i n g o f ob jec ts and regions d i v ided In to na tu ra l or manmade c a t e g o r i e s . E f f o r t s in scene c l a s s i f i c a t i o n to date have concent ra ted on de tec t i on and c l a s s i f i c a t i o n In v ideo imagery. This work w i l l be extended to i n f r a r e d imagery In the near f u t u r e . The goal of a route p lann ing system Is t o analyze a l l a v a i l a b l e in fo rmat ion t o produce a rou te tha t Is opt imal in l i g h t of predetermined mission requ i rements . The TREK rou te p lanning system cons i s t s of t h ree s tages : A) po in t g e n e r a t i o n , B) graph g e n e r a t i o n , and C) h e u r i s t i c search . A. Po in t Generat ion The TREK system generates th ree d i f f e r e n t types of path p o i n t s : two-sweep p o i n t s , convex t e r r a i n po in t s and crossover exposure p o i n t s . The two-sweep po in t genera t ion a lgo r i t hm Is a humanist ic approach to t e r r a i n t r a v e r s a l assuming t h a t the system possesses a map and a compass. The c e n t r a l Idea of two-sweep is the assumption t h a t the d i r e c t i o n of the goal po i n t can always be determined through J. Gilmore and A. Semeco 1087 simple geometry. The a l g o r i t h m I n i t i a l l y scans forward from I t s c u r r e n t l o c a t i o n u n t i l I t locates the goal p o i n t o r encounters an o b s t a c l e . The unde r l y i ng h e u r i s t i c Is the s imple s t r a tegy t h a t a person uses when the re Is an obs tac le In f r o n t t h a t prevents him/her to go to the des i red d e s t i n a t i o n . The I n d i v i d u a l usua l l y goes around the obs tac le f o l l o w i n g I t s contour u n t i l t he d e s t i n a t i o n I s v i s i b l e . This approach can be used both at the loca l and g loba l leve l In r o u t e p l a n n i n g , but on ly the g loba l a p p l i c a t i o n Is discussed In t h i s paper. The convex po in t s of I n t e r e s t f o r an obs tac le are def ined as the se t of p o i n t s on the o b s t a c l e ' s boundary such t h a t : a) each p o i n t In the set Is a convex p o i n t , b) f o r each p o i n t P In the s e t , p 's cu rva tu re Is a local maximum, and c) the set of po in t s forms a convex po lygon . A set of convex p o i n t s Is generated fo r each obs tac le In the t e r r a i n Image. An obs tac le Is represented as a l i s t of ( x , y ) coord ina te p a i r s which de f ine I t s boundary. This set of po in ts does not have to be of a cont inuous n a t u r e , but I t Is assumed t h a t the d is tance between ad jacent po in t s Is u n i f o r m . The combinat ion of two-sweep and convex p o i n t rou tes produces an ex tens ive rou te graph w i th a number of rou te segment I n t e r s e c t i o n s . These I n t e r s e c t i o n s or crossover po in t s a c t u a l l y represent high t r a f f i c areas of maximum exposure. By connect ing a l l crossover po in t s w i th a I I n e o f s I g h t a l g o r i t h m , a number of o f f e n s i v e rou te op t ions are c r e a t e d . When merged w i th defens ive path p o i n t s produced by two-sweep and convex p o i n t g e n e r a t i o n , an unbaslsed rou te graph Is genera ted. In t h i s mode the TREK system Is capable of n o n m i l i t a r y rou te p lann ing as would be the case In a robot v e h i c l e p a t r o l l i n g a n a t i o n a l park look ing fo r s t randed