Inclined Plane Walking Compensation for a Humanoid Robot

Transcription

1 Incline Plne Wlking Compenstion for Humnoi Robot Nttpong Kewlek n Thvi Mneewrn Institute of Fiel Robotics, King Mongkut's University of Technology Thonburi, Bngkok, Thiln (Tel : ; E-mil: k.nttpong@hotmil.co.th, prew@fibo.kmutt.c.th) Abstrct: This pper suggeste the simple compenstion lgorithm for humnoi robot when wlking on n incline plne. Motion trnsition of the robot for vrying slope cn be chieve from justing the center of mss of the robot forwr/bckwr by compensting the robot's nkle. This propose metho use the feebck from the ccelerometer in the robot's boy to etect the inclintion ngle of the groun plne. The propose metho ws teste with smllsize humnoi n showe tht it cn successfully mintin stbility of the wlk on vrying incline plne up to 10 egrees. Keywors: Wlking controller, Humnoi, IMU 1. INTRODUCTION In this pper, we propose how to control humnoi robot to wlk up/own the slope when the slope is chnge. For exmple, Christophe Sbourin et l [1] use the Fuzzy-CMAC to pt the wlking git s function of the informtion on the slope. In their simultion, the robot cn wlk up n own slope when the esire slope is chnging. Lin Yng et l [2] presente three bsic bipel wlking git justment moes which cn be pte to the rough terrin incluing slope on the 2D simultion. Jun Morimoto et l [3] propose moel-bse reinforcement lerning lgorithm to pt the wlking cycle timing to the ynmics of the robot n environment when the robot wlking up/own on the incline plne. Different wlking controllers hve been esigne for humnoi robot such s ZMP-bse control, liner inverte penulum moel control (LIPM). For wlking on n incline plne or n uneven terrin, most stuies use the simultion moel of the robot n its environment with known incline ngle [4] to verify the results or use preictive controller tht inclue externl sensors such s ultrsonic [5] or cmer to sense the terrin he. In this pper we propose the metho tht uses the informtion from the robot internl sensor to just the robot's wlking control lw to cope with chnges of the groun plne incline ngle. The concept of this metho is quite simple n stright forwr. With our existing control lgorithm, the joint position control ws pplie to ll motors on the robot. The blncing control is performe by compensting the joint nkle position with the feebck signl from the gyro sensors. Without the justment tht we re proposing here, when the robot stns on n incline plne, it will try to sty in the upright pose, therefore the boy ngle s references to the worl coorinte irectly represents the ngle of the incline plne. Therefore, by justing the trget position of the nkle so tht it cn properly compenste for the inclintion ngle, the robot will be more stble when wlking on the incline. 2. WALKING CONTROL SYSTEM Fig. 1 The humnoi robot wlking control system Fig.1 shows the blncing control system tht is use in our humnoi robot. The locomotion control lgorithm is performe by the ARM7 microcontroller. After the PC sens locomotion commn to the microcontroller, the joint trjectory for wlking is generte ccoring to the wlking prmeters such s step height, step size, step time n the leg inverse kinemtics. During the wlk cycle, the blncing control is e to the nkle joint control input. The compenstion vlue ws clculte from the robot's ngulr velocity ( b ) tht is sense by the gyroscope ttche insie the robot's boy. In the norml opertion, the nkle control lw cn be represente by (1) + K b Where is the nkle ngle. K is the control gin. (1) /10/$15 ICROS 1403

2 3. ANKLE COMPENSATION FOR INCLINED PLANE WALKING 3.1 Fining nkle compenstion ngle In orer to wlk on n incline, the blncing control lone which only uses the gyro feebck cnnot cope with the chnges of the boy ngle s references to the worl. Therefore the trget ngle for the robot's nkle shoul be juste so tht the boy cn len in the right irection when wlking up/own n incline. In this pper, we propose the nkle compenstion function tht will be use for the incline plne wlking. We use the nkle compenstion function to fin the vlue ( c ) to compenste the robot s nkles. Ours compenstion function ws clculte from the boy ngle tht cn be mesure from the 2-xis ccelerometer ttche insie the robot's boy. In the stning pose, the boy ngle ( b ) cn be clculte from Eq. (2). Fig. 4 The boy ngle n ngulr velocity when the robot ws wlking from 0 to 10 egrees incline with the nkle compenstion ngle of 9.97 egrees. Fig. 5 The boy ngle n ngulr velocity when the robot ws wlking from 0 to 10 egrees incline with the nkle compenstion ngle of egrees. Fig. 2 Boy ngle clcultion from 2-xis ccelerometer b Where A A 1 z tn (2) x x A is the boy ccelertion in X xis. A z is the boy ccelertion in Z xis. However, uring the wlk, the boy ngle hs to be estimte from the ccelerometer signl using moving verge winow filter. After the boy ngle is estimte, the trget ngulr position of the robot's nkle shoul be compenste so tht the center of grvity of the robot's boy cn be move forwr when the robot is wlking up/own the incline plne s shown in Fig. 3. In orer to fin the pproprite compenstion function, we rn n experiment to ientify the proper ngulr bis of the robot's nkle t ifferent incline ngles. The ngulr bis is selecte from experiments s shown in Fig. 4 n 5. The nkle compenstion of is chosen for 10 egrees incline becuse the verge boy ngle of the robot ws mintine in the sme level s t 0 egree, while t 9.97, the verge vlue of boy ngle is lower when the incline ws encountere. The ngulr bis tht the most stble wlking motion cn be relize for ifferent incline ngles were summrize in Tble 1. Fig.6 shows the liner function tht cn be erive from the reltion between the inclintion ngle n the nkle compenstion vlue. Tble 1 The nkle compenstion ngle Fig. 3 Ankle ngle justment on n incline plne /10/$15 ICROS 1404

3 Fig. 6 The liner reltion between the nkle compenstion ngle n the inclintion ngle 3.2 Wlking compenstion From Tble 1, we foun tht the compenstion function for the trget ngulr position of the robot's nkle ccoring to the mesure boy's ngle cn be fitte into liner function s shown in Fig.6. Therefore, fter the boy ngle is estimte from the 2- xis ccelerometer, the new nkle position ( c ) of the robot cn be clculte from the compenstion function s shown in eq.3. Fig. 7 The estimtion of the boy ngle when the inclintion ngle of the plne is chnge () Wlking up from 0 to 10 egrees, (b) Wlking own from -10 to 0 egrees 4. EXPERIMENT The smll-size humnoi robot KM-3 ws use in this experiment. KM-3 hs 18 DOF with 6 DOF in ech leg. This robot is 49 cm tll n weighs 3.3 kg. The 2- xis ccelerometer sensors n the 2 xis gyro sensors were instlle insie the robot's boy. The specifiction of KM-3 is shown in Tble 2. Tble 2 KM-3 Specifiction c + K b (3) Fig. 7 showe the estimtion of the boy ngle from the ccelerometer t tht chnge ue to the chnging incline ngle of the plne. The new nkle position is clculte when the boy ngle estimtion chnges more thn specifie threshol. Normlly, uring the wlk on flt plne, the robot's boy will len forwr bout 7 egrees. We intentionlly set the boy ngle of the robot to len forwr uring tht wlk for the fster wlking spee. In Fig.7 () when the robot wlke from the flt plne to the slope of 10 egrees inclintion, the boy ngle estimtion ws lrgely reuce (t roun time 3 to 4 secons). The new nkle position ws set when the boy ngle estimtion becme lower thn 4 egrees. After the new nkle position ws set ccoring to the nkle compenstion function, the boy ngle increse to the sme level of 7 egrees s before. The sme behvior ws shown in Fig. 7 (b) when the robot wlke own from the incline of -10 egrees to 0 egrees. When the chnge of inclintion ngle ws encountere, the boy ngle estimtion becme lower. The new nkle position ws compenste when the boy ngle estimtion ws lower thn 4 egrees, so tht the boy ngle increse bck to bout 7 egrees. In this experiment, the robot blinly wlke on the incline plne tht chnge its inclintion ngle from 0 egrees to 1-10 egrees n from -10 egrees to -9 until 0 egrees. Two cses were compre, without the nkle compenstion n with the propose nkle compenstion metho. In ech experiment, the robot wlke up n own 10 trils. The loction of the incline chnge is rnomly chosen t ech tril, thus the robot cn only sense the loction of the inclintion chnges from the chnges of its boy ngle estimtion using the ccelerometer /10/$15 ICROS 1405

4 Fig. 9 Sequence of successive frmes extrcte from the vieo showing the robot when wlking up on the incline plne tht chnge from 0 to 10 egrees successfully. Fig. 8 The experimentl humnoi robot 4.1 Wlking up n incline In this experiment, the robot woul wlk up on the flt plne (0 egree) into the incline plne tht hs 10 egrees inclintion ngle using the compenstion function tht we foun from Tble 1. The robot's boy ngle ws set to len forwr t 7 egrees uring the wlk. The wlking prmeters were set s shown in Tble 3. The new nkle ngle ws juste when the estimte boy ngle from the ccelerometers ws lower thn 4 egrees. Tble 3 Wlking prmeters The experimentl results show tht with the propose nkle compenstion bse on the boy ngle estimtion from ccelerometer sensor, the robot cn successfully wlk up the incline plne tht chnge from 0 to 10 egrees. Fig.9 showe the frme sequence of the robot tht cn wlk up on the incline plne tht chnge from 0 to 10 n Fig. 10 showe the comprison of the boy ngle estimtion of the robot with n without nkle compenstion. Without nkle compenstion, the robot fell bckwr fter the inclintion chnge ws encountere. Fig. 10 The boy ngle estimtion of the robot when wlking on the chnging slope from 0 to 10 egrees. () Without compenstion, (b) With nkle compenstion 4.2 Wlking own n incline After we successfully pplie the nkle compenstion control to mke the robot wlk from the flt plne up to the incline plne. We lso use this propose metho to control the robot to wlk ownwr from the incline plne tht hs -10 egrees inclintion ngle to the flt plne. However, when the robot is wlking own on the incline plne, it hs higher ccelertion thn wlking up. Therefore, the robot s wlking velocity h to be reuce by justing the step size. We ecrese the step size from to 0.02 meters/step in the wlking own experiment. Other wlking prmeters were kept to be the sme s in the wlking up experiment s shown in Tble 4. The experimentl result ws shown in Fig. 11~ /10/$15 ICROS 1406

5 ngle chnges from 0 to 10 egrees n from -10 egrees to 0 egrees successfully using the propose nkle compenstion metho. However, since the compenstion nkle justment cn only be clculte from the boy ngle mesurement from the ccelerometers, this metho woul not be sufficiently robust to cope with lrge chnges of inclintion ngle of more thn 10 egrees. Aitionl sensors tht cn provie preictive informtion of the terrin he might be useful when the chnges of inclintion ngle re lrge. Fig. 11 Sequence of successive frmes extrcte from the vieo showing the robot when wlking up on the incline plne tht chnge from -10 to 0 egrees successfully. Fig. 12 The boy ngle estimtion of the robot when wlking on the chnging slope from -10 to 0 egrees. () Without compenstion, (b) With nkle compenstion REFERENCES [1] C. Sbourin, K. Mni n O. Bruneu, Autonomous bipe git pttern bse on Fuzzy- CMAC neurl networks, Integrte Computer- Aie Engineering 14, [2] L. Yng, C. M. Chew n A. N. Poo, Rel-time Bipel Wlking Ajustment Moes using Truncte Fourier Series Formultion, IEEE-RAS Interntionl Conference on Humnoi Robots, Pittsburgh, Pennsylvni, USA, [3] J. Morimoto, G. Cheng, C. G. Atkeson n G. Zeglin, A Simple Reinforcement Lerning Algorithm for Bipe Wlking, Interntionl Conference on Robotics n Automtion, [4] C. M. Chew, Blin Wlking of Plnr Bipe on Slope Terrin, Mster Thesis, MIT [5] S. Kjit n K. Tnie, Aptive Git Control of Bipe Robot Bse on Reltime Sensing of the Groun Profile, Autonomous Robots, pp , Without the propose compenstion metho, the robot cnnot mintin stbility when the inclintion ngle of the plne chnges. This propose compenstion metho is very simple n cn be compute in rel-time using the on-bor microcontroller. This metho oes not require itionl sensors such s force sensor or cmer. However, the propose metho is quite limite in its lck of preictbility of the ction he becuse the compenstion cn only ctivte bse on the chnge of its boy ngle fter the robot is lrey on the incline plne. 5. CONCLUSION In this work, the compenstion control lw for the humnoi robot for incline plne wlking is propose. By estimting the boy ngle of the robot from the 2- xis ccelertion sensors which were processe by the moving-verge winow filter, the new trget position of the nkle ws compenste bse on the compenstion function erive from the experimentl result in Tble 1. From the experiment, the humnoi robot cn wlk up/own the slope with the inclintion /10/$15 ICROS 1407