Human Friendly Autonomous Robot using Dempster-Shafer Sensor Fusion and Velocity Potential Field Control

Transcription

1 ICAS 205 : Te Elevent Interntionl Conference on Autonomic n Autonomous Systems Humn Frienly Autonomous Robot using Dempster-Sfer Sensor Fusion n elocity Potentil Fiel Control Dn-Sorin Necsulescu, Yu Hu, Deprtment of Mecnicl Engineering, University of Ottw, Ottw, Cn e-mil: necsu@uottw.c, yu06@uottw.c Jurek Ssiek Deprtment of Mecnicl n Aerospce Engineering, Crleton University, Ottw, Cn, e-mil: jurek_ssiek@crleton.c Abstrct In tis pper, umn frienly utonomous robot, ws presente. Nvigtion of tis robot pplies sensor fusion tecnique bse on Dempster-Sfer meto n velocity potentil fiel. Te controller of te utonomous mobile robot is esigne to le te robot towr te gol wile voiing collisions in complex environments n mintin umn sfety s first priority. Te pproc is bse on Dempster-Sfer sensor fusion of signls from sonr n pssive infrre sensors to llow te robot to ientify umn presence. A velocity potentil fiel robot controller is formulte for te cse of voiing collisions wile giving iger priority to collision voince wit umns s oppose of objects. Simultions n experiments illustrte te performnce of te pproc in extreme situtions. Keywors- umn frienly robots; Dempster-Sfer evience teory; sensor fusion; umn sfety. I. INTRODUCTION Recently, wit rise of te cost of lbor, robots ply more importnt roles tn before. To become more wiely pplie in te rel worl, it is necessry tt te mobile robots ve muc more rtificil intelligence to work in more complex environments, wic migt inclue umns or veicles wit unknown n unpreictble motions, suc s fctory or ome environments. To cieve tis gol, te robot s to be ble to ientify n istinguis te tcles, n te umn n voi collisions wit bot of tem. Furtermore, te robot soul lwys put umn sfety s its first consiertion. In tis pper, sensor fusion pproc bse on Dempster-Sfer evience teory is combine wit velocity potentil fiel pproc use for robot control. Te justifiction of tis coice is te bility of Dempster-Sfer evience teory to inclue supporting evience, refuting evience n n uncertinty intervl, tt permit suitble use of expert knowlege regring utonomous robots nvigtion issues. Te etection re of te robot is subivie into severl zones. Te sensor fusion pproc is employe to fuse bot uncertin ervtions of sonr sensors n pssive infrre sensors n to estimte te probbility of being occupie by umn in ec zone [] [2]. Ten, umn frienly mobile robot nvigtion pproc is use for robot motion control. Te controller is bse on te velocity potentil fiel meto (PF), wic is use to le te robot moving to te gol voiing te tcles [3]. For te novel controller, in tis cse n improvement of te PF llows te robot to voi te umn first priority n only fterwrs te tcles. Tis novel PF pproc ws not pplie before for robot collision voince. Dempster-Sfer evience teory permits to fuse outputs from vrious sensors n ten provie te PF controller wit te require istnces between te robot, te tcles n te gol. Te pper focuses on te novel results presente in te tesis [4]. Te pper presents in Section 2 te Demster Sfer evience teory use for sensor fusion. Section 3 focuses on te moel use for te clcultion of te voince istnce to tcles. Section 4 presents velocity potentil fiel pproc for robot nvigtion in te presence of umns. Simultion results re te topic of section 5, wile experimentl results re presente in Section 6, followe by conclusions in Section 7. II. SENSOR FUSION METHOD BASED ON DEMPSTER- SHAFER EIDENCE THEORY In tis pper, two types of sensors re use to elp te robot sense te environment. Te first one is te Pssive Infrre Sensor (PIR sensor), wic senses te et emitte by umns. PIR sensor is, owever, not sufficiently ccurte. Performing unres of experiments, in t bout 0 % of te experiments te umn sensor i not work well. Besies, PIR sensor migt ientify wrm ir generte by et source s coming from umn. Te secon type of sensor is te ultrsonic sensor. It emits ig frequency soun wves to te objects n ten receives tem to etermine ow fr tey re. Since umn is te first priority of te robot, informtion wic is collecte by ifferent kins of sensors is combine to ientify umn wit iger probbility [4]-[9]. Dempster-Sfer evientil teory (D-S teory) ws cosen to support te probbility clcultion given its bility wen te sensors contributing informtion cnnot ssocite 00 % probbility to teir output ecisions. Te lgoritm cptures n combines wtever certinty exists in te Copyrigt (c) IARIA, 205. ISBN:

2 ICAS 205 : Te Elevent Interntionl Conference on Autonomic n Autonomous Systems object-iscrimintion cpbility of te sensors. Knowlege from multiple sensors bout events (clle propositions) is combine using D-S teory to fin te intersection or conjunction of te propositions n teir ssocite probbilities [4], using m( C k ) A i Bj C ; C 0 A B 0 i k j k m( A ) m( B ) m( A ) m( B ) were A i n B j re te focl elements of m A n m B, respectively. m A n m B re te Bsic Probbilities Assignments (BPA) wic re combine wile C k re te focl elements of te combine BPA [2]. Te frme of iscernments is in tis cse set of cells in te occupncy gris s sown in Figure. Te eigt gris re ble to cover te umn sensors etection re. i i j j () te ll progress. 7. Repet from step one gin until te mission is complete. III. CALCULATION OF THE AOIDING DISTANCE Te robot, te umn zone n tcle zone, sown in Figure 2, were built roun te robot to voi collision even in te worst cse wen te robot s te lest mount of time to voi e-on collision. We enote te umn zone from Figure 2 s A 0 te tcle zone s A n te work re excluing A 0 n A s A. Te spes of te zones were esigne bse on te positions of te sensors. Figure 2. Te Humn Zone n te Obstcle Zone Figure. Humn Sensors Detection Are Bot ultrsonic sensors n PIR sensors keep wtcing n ec sensor contributes by ssigning its BPA over its own frme of iscernments [0] []. Te D-S teory is use to combine two focl elements BPAs. Sensor fusion is n itertive process wit time step cosen of 0.05s, escribe s follows:. Re te ultrsonic sensor outputs looking for te presence of umn. (ssign te BPAs to ec gri) 2. Re te umn sensors outputs.(ssign te BPAs to ec gri) 3. Combine te finl results of bot kins of sensors using D-S teory fter one time step. 4. Clculte te probbility (te combine BPAs) tt quntifies ow probble te gri is to be occupie by umn. 5. Compre te probbility vlue wit te setting vlue n mke te ecision weter tere is umn in te gri n ten generte corresponing response. 6. Initilize te BPA of te occupie gri to zero to restrt Te umn zone ws esigne for umn voince. In te worst umn voince cse, te umn n te robot were moving towr ec oter. Once te robot etecte te umn in its umn zone, it turns into ifferent irection, from te originl irection towr te gol, to voi te collision. Given te umn size n velocity of te umn n te robot te clcultion is bse on r 0. 4m (2) mry mr ( t t ) tr (3) mr t mry 2 ( t t) mryt mry r (4) 2 were is te mximum velocity of te robot, t is te time tt robot nees to ccelerte from zero velocity to mr in Y-irection, mry is te istnce tt te robot is moving in Y-irection, is te rius of te robot, mry is te robot Copyrigt (c) IARIA, 205. ISBN:

3 ICAS 205 : Te Elevent Interntionl Conference on Autonomic n Autonomous Systems ccelertion mgnitue in Y-irection, t is te totl time of te voince n r is te rius of umn ctive re. Finlly, te rius of te umn zone r 0 soul stisfy te constrint r x (5) 0 mrx were x n mrx is te umn n robot moving istnce in X-irection, respectively. Bse on te bove moel, clcultion for te worst cse, gives te rius of te umn zone of 30cm. I. ELOCITY POTENTIAL FIELD METHOD WITH THE CONSIDERATION OF HUMAN PRESENCE In tis pper, te velocity potentil fiel meto [3], obtine from velocity potentils efine in yroynmics, is moifie for plnning pt wic vois collisions wit te umn n te tcles, suc tt, uring te voince, te robot will voi te umn before it strts to voi te tcles. Te robot is guie by its velocity commns wic re given by its nvigtion controller (Figure 3). g r 2 mr ( ) (7) were mr is te mximum velocity of robot. Te repulsive velocity function is efine s [3] rep gol r 0.5 (8) were is te istnce between te tcle n te robot wic is obtine by te sensors. Here, we lso crete repulsive velocity function for te umn, wic s lrger gin tn te repulsive velocity function for tcle Bot repumn r 0.7 (9) n r in te eqution cuse sustine increse of te repulsive velocity if te robot keeps pprocing te tcles or umn, wic mens tt te closer te tcle or umn is, te lrger is te repulsive velocity. Te rottion velocity function is efine s [4] rot r 0.5 (0) were is te istnce between tcle n robot wic is obtine by te sensors. Figure 3. Te Attrctive, Rottion n Repulsive elocity Commns Te istnce between robot n gol gol is 2 2 gol ( yg y ) ( xg x) (6) were refers to te gol n to te center n robot. Te gol rius is g r 0. 4m PF pproc, presente in etil in [3], results in velocity commns for te robot controller. Te ttrctive velocity function is efine s [3] Te resultnt velocity commn is given by, Pr A rep rep rep rot rot rot P A, P A P A, P A, P A P A, P r r sum () r 0 0 A were,, rep, n rot, re te ttrctive, repulsive n rottion velocity commns, respectively. Te rep, n rot re velocity commns regring umn. A is te wole mp re wile A 0 n A represent te umn zone n tcle zone. Copyrigt (c) IARIA, 205. ISBN:

4 ICAS 205 : Te Elevent Interntionl Conference on Autonomic n Autonomous Systems. SIMULATION RESULTS FOR HUMAN FRIENDLY ROBOT In orer to test if te robot is ble to voi collisions s expecte, simultions were crrie out using MATLAB TM. Figure 4 presents te results of te simultion wit te following symbols: Te blue rectngle wit te re circle: te umn Te blck rectngle: te tcle Te blue polygon wit ligt blue circle: te robot In tese simultions, te robot i not know it will fce ngerous sitution. Te im of creting suc n extreme sitution is to test if te robot controller s te bility to voi te umn n te tcle wile it still s time n spce. (c) () () (e) (b) (f) Copyrigt (c) IARIA, 205. ISBN:

5 ICAS 205 : Te Elevent Interntionl Conference on Autonomic n Autonomous Systems to te snpsots of experimentl results from Figure 5 ()- (f). Experimentl results confirme tt te propose combintion of Dempster-Sfer sensor fusion wit velocity potentil fiel bse controller successfully voie fixe tcles n umns in moving towrs te esignte gol. (g) () () Figure 4. Higer Humn Priority (Two Humns n One Obstcle) As sown in Figure 4, te robot turne roun n rn wy from te two umn n, fterwrs voie te tcle moving in irection were tere is no nger. Tis illustrte tt te propose PF bse controller permits to voi fixe tcles n voi umns n rrive to te esignte gol. (b) I. EXPERIMENTAL RESULTS Experiments were crrie out for te sme scenrio s in simultions, in wic te robot suenly fces te moving umn n tcles t te sme time n, uring te voince, te robot soul etour umn n tcles wit ifferent priorities, suc tt te umn lwys s iger priority in voiing collision tn tcles. For tis scenrio, two umns kept pprocing te robot locte in front n left sie n te wll ws in its rigt sie. Te purpose to esign tis sitution is to test if te robot is ble to voi umns first (even s risk to collie wit n tcle) n only fterwrs to strt voiing te tcles. As we cn see from te Figure 5, te robot voie te two umns first even wile moving in te irection of te wll. Ten, te robot voie te wll n rn wy from te nger of collision. It cn be note tt te simultion results from Figure 4 ()-(f) correspon (c) Copyrigt (c) IARIA, 205. ISBN:

6 ICAS 205 : Te Elevent Interntionl Conference on Autonomic n Autonomous Systems () on its own. Tis meto is goo enoug for umn etection, but it cnnot ensure complete security for te umn. In te future, noter sensor bse on infrre cmer will ve to be e to te robot so tt umn movements cn be furter recore by te robot troug processing pictures cpture by te infrre cmer n tus te umn sfety cn be furter improve. (e) (f) Figure 5 Two Humn n One Obstcle Avoince II. CONCLUSIONS Te pper presents novel pproc to robot-umn collision voince using Dempster-Sfer meto bse on evience teory. Tis metoology llows to integrte signls from multiple sensors. Tis supplies more relible istnce estimtion to velocity potentil fiel controller. Te results presente in tis pper sow tt te new umn frienly mobile robot nvigtion controller bse on Dmpter-Sfer sensor fusion is ble to le te robot, voi collisions wit te tcles n umns wile lwys mintining umn sfety s its first priority. In te extreme cse, wen te robot oes not ve enoug room to voi te collision wit te umn n n tcle, it cooses to protect te umn even if in te process it migt collie wit n tcle. In te work presente in tis pper, te robot s only one infrre sensor to sense te umn. Altoug te sonr sensor elps te infrre sensor it cnnot istinguis umn REFERENCES [] G. Benet, F. Blnes, J. E. Simó, n P. Pérez, Using infrre sensors for istnce mesurement in mobile robots, Rob. Auton. Syst., vol. 40, no. 4, Sep. 2002, pp [2] G. Tun,. C. Gungor, n K. Gulez, An utonomous wireless sensor network eployment system using mobile robots for umn existence etection in cse of issters, A Hoc Networks, vol. 3, Feb. 204, pp [3] E. Pruner, Control of Self-Orgnizing n Geometric Formtions, A tesis presente for te egree of Mster of Applie Science in engineering, University of Ottw, 203, pp [4] L. A. Klein, Sensor n t fusion: tool for informtion ssessment n ecision mking, Spie.org, 2004, pp [5] R. R. Murpy, Dempster Sfer Teory for Sensor Fusion in Autonomous Mobile Robots, IEEE Trnsctions on Robotics n Automtion, vol.4, No.2, Apr 998, pp [6] L. Zeng, Design n control of umn frienly robots, A tesis presente for te egree of P.D of Applie Science in engineering, McMster University, 200, pp [7] H. M. Coset, Principles of robot motion: teory, lgoritms, n implementtions. MIT press, 2005, pp [8] G. M. Bone n L. Zeng, Collision voince for nonolonomic mobile robots mong unpreictble ynmic tcles incluing umns, 200 IEEE Int. Conf. Autom. Sci. Eng, Aug. 200, pp [9] Y. Lu, L. Zeng n G. M. Bone, Multisensor System for Sfer Humn-Robot Interction, 2005 IEEE Interntionl Conference, April, 2005, pp [0] G. Feng, X. Guo n G. Wng, Infrre motion sensing system for umn-following robots, Sensors Actutors A Pys., vol. 85, Oct. 202, pp. 7. [] M. Km, X. Zu n P. Klt, Sensor Fusion for Mobile Robot Nvigtion, Proceeings of te IEEE 997, vol. 85, no., 997, pp08-9. [2] T. Ali n P. Dutt, Metos to Obtin Bsic Probbility Assignment in Evience Teory, Int. J. of Computer Applictions, Jn. 202, pp [3] Y. Guo, A. Song, J. Bo, H. Tng n J. Cui, A Combintion of Terrin Preiction n Correction for Serc n Rescue Nvigtion, Int. J.of Avnce Robotic Systems, No. 3, 2009, pp [4] Y. Hu, Humn Frienly Robot, MASc Tesis, University of Ottw, 204, pp Copyrigt (c) IARIA, 205. ISBN: