Autonomous urban reconnaissance using man-portable UGVs

Size: px

Start display at page:

Download "Autonomous urban reconnaissance using man-portable UGVs"

Holly Mathews
5 years ago
Views:

1 Auonomous urban reconnaissance using man-porable UGVs Brian Yamauchi irobo Corporaion, 63 Souh Avenue, Burlingon, MA ABSTRACT For he Wayfarer Proec, funded by he US Army hrough TARDEC, we have developed echnologies ha enable manporable PackBo Wayfarer UGVs o perform auonomous reconnaissance in urban errain. Each Wayfarer UGV can auonomously follow urban srees and building perimeers while avoiding obsacles and building a map of he errain. Each UGV is equipped wih a 3D sereo vision sysem, a 360-degree planar LIDAR, GPS, INS, compass, and odomery. The Hough ransform is applied o LIDAR range daa o deec building walls for sree following and perimeer following. We have demonsraed Wayfarer s abiliy o auonomously follow roads in urban and rural environmens, while building a map of he surrounding errain. Recenly, we have developed a ruggedized version of he Wayfarer Navigaion Payload for use in rough errain and all-weaher condiions. The new payload incorporaes a compac Tyzx G2 sereo vision module and a high-performance Ahena Guidesar INS/GPS uni. 1. INTRODUCTION We have developed urban navigaion capabiliies for he man-porable irobo PackBo as par of he Wayfarer Proec funded by he US Army Tank-Auomoive Research, Developmen, and Engineering Cener (TARDEC). The Wayfarer PackBos are designed o perform fully-auonomous reconnaissance missions in urban errain, reducing he risk o warfighers by keeping hem ou of harm s way. Each Wayfarer PackBo is equipped wih a SICK LD OEM 360-degree planar LIDAR, a sereo vision sysem, an INS/GPS, compass, and odomery. In previous papers, we described he Scaled Vecor Field Hisogram (SVFH) obsacle avoidance echnique 5 we developed for Wayfarer (an exension of Borensein and Koren s Vecor Field Hisogram 2 ), and he auonomous mapping capabiliies ha we inegraed wih auonomous perimeer-following behaviors 6. In combinaion, hese echniques enabled Wayfarer o perform fully-auonomous perimeer reconnaissance missions. In his paper, we repor our recen progress in developing auonomous roue reconnaissance capabiliies using sreefollowing behaviors. We also describe a ruggedized version of he Wayfarer Navigaion Payload ha we have developed for use in rough errain and all-weaher condiions. 2. ROUTE RECONNAISSANCE 2.1. Sree-Following Behavior The Wayfarer follow-sree behavior uses a Hough ransform 1 o deec linear feaures in he 360-degree planar LIDAR range daa. Each line is hen classified based on wheher i is on he righ or side of he robo: if θ min < θl < θ side( L) = righ if θrigh min < θl < θ none oherwise max righ max where L is a line, side(l) is he side of line L, θ L is he orienaion of line L, θ min and θ max bracke he region of ineres on he side and θ righ min and θ righ max do he same for he righ side. Currenly θ min = 0, θ max = θ righ min = π, and θ righ max = 2π, so all lines excep hose orhogonal o he robo s curren heading are classified as being on he or righ. Copyrigh 2006 Sociey of Phoo-Opical Insrumenaion Engineers. This paper will be published in he Proceedings of Unmanned Sysems Technology VIII, SPIE Conference 6230, and is made available as an elecronic preprin wih permission of SPIE. One prin or elecronic copy may be made for personal use only. Sysemaic or muliple reproducion, disribuion o muliple locaions via elecronic or oher means, duplicaion of any maerial in his paper for a fee or for commercial purposes, or modificaion of he conen of he paper are prohibied.

2 The line headings are used o updae separae accumulaor arrays for he and righ sides of he robo. The accumulaor arrays filer ou ransien lines generaed by he Hough ransform and produce a more sable desired heading. The value of he accumulaor bins a ime is given by: a a, righ, = = ( 1 λ) a, 1 + λ v( H ) : iβ < θ ( H ) < ( i+ 1) β, side( H ) ( 1 λ) arigh, 1 + λ v( H ) : iβ < θ ( H ) < ( i+ 1) β, side( H ) where a, i,-1 is he accumulaor bin value a he previous imesep, a righ, i,-1 is he righ accumulaor bin value a he previous imesep, λ is he decay rae (beween 0 and 1), H is he se of lines deeced by he Hough ransform, H is he h line from his se, v(h ) is he number of poins voing for his line, θ(h) is he orienaion of he line, and β is he bin size. Noe ha all of hese orienaions are in world coordinaes, no robo-relaive coordinaes. The seleced heading corresponding o he maximum bin in each accumulaor is given by: θ = ( i + 0.5) β : : a, a,, = = righ θ righ = ( i + 0.5) β : : arigh, arigh,, The behavior hen compues he average of he and righ headings as defined by: θ desired = θ + ( θ, θ ) 2 righ ( θ, θ ) righ θ = θ righ θ θ righ if π < θ righ θ oherwise π Follow-sree hen sends θ desired as he desired heading o he SVFH obsacle avoidance behavior 5.

3 Figure 1: Line pair deeced by follow-sree a hallway inersecion Figure 1 shows he follow-sree behavior operaing indoors in a hallway. The blue lines indicae he line pair (sronges lines on and righ sides) deeced by he behavior. The yellow line indicaes he desired heading compued as he average of he maximum and righ accumulaor values Mixed Urban/Rural Sree Following We esed Wayfarer s sree following sofware in an area of mixed urban and rural errain near irobo Headquarers in Burlingon, Massachuses. The oudoor es environmen consiss of an unmarked road ha runs hrough a wooded area and hen behind an urban facory. The oal lengh of his road was approximaely 200 meers each way. During hese rials, he robo was able o successfully follow roads a speeds up o 2 meers/second (4.5 mph), he maximum speed of he PackBo plaform. This is paricularly noable given he vas difference beween he irregular ouline and exure of he wooded ree line and he fla walls previously esed during sree and perimeer following. Figure 2 shows Wayfarer auonomously following he rural secion of his road uphill. Figure 3 shows Wayfarer auonomously navigaing around a fallen ree as i follows he road. SVFH obsacle avoidance is seamlessly inegraed wih he sree following behavior. The follow-sree behavior provides a desired heading, and he avoid-obsacles behavior seers he robo owards he clear heading closes o he desired heading. Afer he robo seers around he obsacle, avoid-obsacles smoohly guides he robo back o following he sree heading provided by follow-sree. Figure 4 shows Wayfarer auomaically ransiioning from following he rural secion of his road o following he urban secion behind he facory. A key feaure of our behavior-based approach o navigaion is how i enables robus navigaion in very differen environmens. Wayfarer uses exacly he same behaviors o follow urban srees as o follow rural roads. No parameers need o be changed and no user inervenion is required as Wayfarer follows he road from a rural o an urban environmen.

4 Figure 2: Wayfarer auonomously following rural road downhill Figure 3: Wayfarer auonomously navigaing around fallen ree

5 Figure 4: Wayfarer auomaically ransiioning from rural o urban sree following Figure 5: Map of urban/rural errain buil by Wayfarer (10 meer grid spacing)

RUGGEDIZED WAYFARER NAVIGATION PAYLOAD Figure 6: Wayfarer PackBo wih ruggedized navigaion payload We have recenly developed a ruggedized payload housing (Figure 6) for he Wayfarer sensors.

6 Figure 5 shows he occupancy grid map 3 buil by Wayfarer as i auonomously followed he road uphill hrough rural and urban errain. Wayfarer sared a he righ edge of his map (rural) and followed he road uphill o he, where i emerged in he urban errain behind he facory Tyzx G2 Sereo Vision Sysem 3. RUGGEDIZED WAYFARER NAVIGATION PAYLOAD Figure 6: Wayfarer PackBo wih ruggedized navigaion payload We have recenly developed a ruggedized payload housing (Figure 6) for he Wayfarer sensors. This payload inegraes a Tyzx G2 Sereo Vision Module, a SICK LD OEM LIDAR, and an Ahena Guidesar GS-111m INS/GPS in a rainproof housing. The housing includes an inegraed rain shield/roll plae above he SICK LIDAR, and a rubber sheah around he lower LIDAR housing ha prevens waer from seeping ino he payload. While his payload is no immersible, i is suiable for use in rain or snow. We replaced he earlier prooype s Poin Grey Bumblebee sereo vision sysem and Plug-N-Run vision CPU wih a single Tyzx G2 sereo vision module. The G2 provides 512 x 320 deph images a 30 Hz, using a cusom ASIC chip along wih a COTS Texas Insrumens DSP o perform sereo image processing. The G2 module is smaller he sandard PackBo Scou head and is mouned o a rugged upper meal bracke wihin he Scou head volume. The G2 module uses is lower surface as a hea sink, so we mouned he sensor wih a space beween he G2 underbody and he PackBo Scou chassis. This allows he G2 o dissipae hea effecively. The G2 is fully-inegraed wih he Wayfarer hardware and sofware, providing 3D obsacle avoidance capabiliies for Wayfarer. To increase speed, he obsacle deecion sofware focuses on a horizonal region 100-pixels high near he cener of he image, roughly corresponding o obsacles o obsacles ha are oo all o climb bu oo low o pass under.

7 Figure 7: Close-up for Tyzx G2 Module in ruggedized Wayfarer payload Figure 8: Opical pairs and deph image from Tyzx sereo vision sysem oudoors

To reduce spurious false posiives in he obsacle map, coniguous regions in he image are grouped ino blobs. Each blob consiss of a se of pixels ha are adacen in he image and close o one anoher in range.

We found ha his approach significanly reduces he level of noise in he range image and subsanially reduces he number of false posiives in he obsacle map.

8 To reduce spurious false posiives in he obsacle map, coniguous regions in he image are grouped ino blobs. Each blob consiss of a se of pixels ha are adacen in he image and close o one anoher in range. Blobs smaller han a specified hreshold are removed from he image. We found ha his approach significanly reduces he level of noise in he range image and subsanially reduces he number of false posiives in he obsacle map. Figure 8 shows he opical pairs and deph image from he Tyzx DeepSea sereo vision sysem ouside irobo Headquarers. G2 performance is similar. This image shows ha he vision sysem can accuraely deec a wide range of naural and arificial obsacles, including rees, cars, poles, and buildings. This also shows ha he sysem can accuraely deermine he range o he ground plane, wheher naural (grass) or arificial (asphal, concree) Ahena GuideSar INS/GPS Sysem We worked wih Ahena Technologies of Warrenon, Virginia o develop an improved localizaion payload for Wayfarer. This payload is a cusomized version of he Ahena Guidesar GS-111m INS/GPS uni ha incorporaes he PackBo moion model and odomery daa. By incorporaing he PackBo s moion model and odomery ogeher wih INS and GPS in he Kalman filer, he Guidesar provides improved localizaion accuracy in areas wih and wihou GPS coverage. We have fully inegraed he Guidesar wih he Wayfarer mapping sysem. Our resuls have shown a subsanial increase in localizaion accuracy using his sysem, especially over large areas and long periods of ime. Figure 9: irobo HQ exerior map generaed using raw odomery () and Ahena Guidesar (righ) (10 meer grid spacing)

9 Figure 10: irobo HQ exerior and parking lo map generaed using Guidesar (10 meer grid spacing) Figure 11: Close-up of saring and ending poins for HQ exerior and parking lo map (1 meer grid spacing)

Figure 12: irobo HQ exerior and parial inerior map generaed using Guidesar (10 meer grid spacing) Figure 9 () shows an (indecipherable) map of he irobo HQ building using raw odomery.

10 Figure 12: irobo HQ exerior and parial inerior map generaed using Guidesar (10 meer grid spacing) Figure 9 () shows an (indecipherable) map of he irobo HQ building using raw odomery. In conras, Figure 9 (righ) shows a map of he exerior of irobo HQ generaed using he Guidesar. This map was generaed by eleoperaing he robo while he mapper auomaically consruced a map of he robo s environmen using he SICK LIDAR. The green line shows he pah aken by he robo. The whie areas represen obsacles; he black areas represen open space; and he gray areas represen unknown erriory. The blue lines are spaced a 10 meer inervals. The oal disance raversed by he robo is approximaely 500 meers. A he end of his run, we drove he robo back o is saring posiion, and is measured posiion was wihin 2 meers of is acual posiion. Figure 10 shows a map of boh he irobo HQ exerior and he adacen parking los generaed by he mapper using he Guidesar for localizaion. The ime required o generae his map was approximaely 45 minues. The oal disance

11 raversed was over 1 kilomeer, and he area mapped covers roughly 25,000 square meers. A he end of his run, we drove he robo back o is iniial posiion, and is measured posiion was wihin 4 meers of is acual posiion. Figure 11 shows a close-up (1 meer grid spacing) of he saring posiion of he robo (START), and he measured posiion when he robo was moved back o he saring posiion a he end of he run (FINISH). Afer mapping he HQ exerior and parking los, we drove he Wayfarer PackBo inside he HQ building, and he Guidesar and he mapper auomaically ransiioned o localizaion based on INS and odomery alone (wihou GPS). Figure 12 shows a parial map of he inerior hallways of he HQ building generaed as we drove he PackBo from he doorway o my office. 4. CONCLUSIONS We have developed auonomous roue reconnaissance capabiliies for he Wayfarer PackBo. Wayfarer is able o follow boh urban and rural roads, as long as road-aligned feaures exis on a leas one side of he road. As Wayfarer explores he roue, i auomaically builds an occupancy grid map of is surroundings. We have also developed a ruggedized version of he Wayfarer Navigaion Payload. While previous experimens were conduced wih a loosely-inegraed prooype, he new payload is designed for use in rough errain and in all-weaher condiions. The rugged payload includes a compac Tyzx G2 sereo vision module and a high-performance Ahena Guidesar GS-111m INS/GPS sysem. We have inegraed he G2 wih our obsacle avoidance sysem, and we have inegraed he Guidesar wih our mapping sysem. In fuure work, we inend o ransiion he Wayfarer Navigaion Payload o a produc for he irobo PackBo produc line, providing fully-auonomous navigaion capabiliies for man-porable robos in urban errain. 5. ACKNOWLEDGMENTS This research was funded by he US Army TARDEC (DAAE07-03-C-L147). Rainer Gasche and Erik Schoenfeld inegraed he elecrical and mechanical componens of he Wayfarer PackBos. Todd Pack developed he Bumblebee sereo vision sofware used for obsacle avoidance. Dan Grollman wroe he Hough ransform sofware. Boris Sofman wroe he Tyzx obsacle avoidance sofware. Richard Juchniewicz designed he ruggedized payload housing. We hank Tyzx and Ahena for heir suppor wih payload inegraion. REFERENCES 1. D. Ballard and C. Brown, Compuer Vision. Upper Saddle River, New Jersey: Prenice-Hall, J. Borensein and Y. Koren, "Real-ime obsacle avoidance for fas mobile robos." IEEE Transacions on Sysems, Man, and Cyberneics, Vol. 19, No. 5, Sep./Oc. 1989, pp H. Moravec and A. Elfes, High resoluion maps from wide angle sonar. In Proceedings of he 1985 IEEE Inernaional Conference on Roboics and Auomaion, pp B. Schiele and J. Crowley, A comparison of posiion esimaion echniques using occupancy grids. In Proceedings of he 1994 IEEE Inernaional Conference on Roboics and Auomaion, May B. Yamauch The Wayfarer modular navigaion payload for inelligen robo infrasrucure. In Unmanned Ground Vehicle Technology VII, edied by G. Gerhar, C. Shoemaker, and D. Gage, Proceedings of SPIE Volume 5804 (SPIE, Bellingham, WA, 2005), pp B. Yamauch Wayfarer: an auonomous navigaion payload for he PackBo. In Proceedings of AUVSI Unmanned Vehicles Norh America 2005, Balimore, MD, June 2005.

Spring Localization I. Roland Siegwart, Margarita Chli, Martin Rufli. ASL Autonomous Systems Lab. Autonomous Mobile Robots

Spring Localization I. Roland Siegwart, Margarita Chli, Martin Rufli. ASL Autonomous Systems Lab. Autonomous Mobile Robots Spring 2017 Localizaion I Localizaion I 10.04.2017 1 2 ASL Auonomous Sysems Lab knowledge, daa base mission commands Localizaion Map Building environmen model local map posiion global map Cogniion Pah