2008 8th IEEE-RAS International Conference on Humanoid Robots December 1 ~ 3, 2008 / Daejeon, Korea A Friction Based “Twirl” for Biped Robots Kanako Miura 1 , Shin’ichiro Nakaoka 1 , Mitsuharu Morisawa 1 , Kensuke Harada 1 , Shuuji Kajita 1 1 National Institute of Advanced Industrial Science and Technology (AIST), Japan [email protected], [email protected] [email protected] [email protected], [email protected] Abstract— This paper presents preliminary results on generating turning motion of a humanoid robot by slipping the feet on the ground. We start by presenting the fact that such slip motion is used by humans based on actual human motion capture data, and show how this is necessary for sophisticated humanlike motion. To generate the slip motion, we predict the amount of slip using the hypothesis that the turning motion is caused by the effect of minimizing the power generated by floor friction. Verification is conducted through both simulation and experiment with the humanoid robot HRP-2. A ”twirl” utilizing both feet slip is successfully demonstrated. I. I NTRODUCTION In our daily life, we unconsciously exploit the fact that our feet slip on the ground, when we turn or stop. However, most current humanoid robot locomotion assumes no slip contacts. As a result, the robots tend to make many small steps when turning in place, which usually takes a lot of time to complete. Taking all these steps does not seem effective in terms of both energy consumption and stability. Thus we believe the use of slip is important to apply quick and smooth robot motion. This paper is a preliminary discussion of the slip phenomenon, aiming to realize highly sophisticated human like motion on humanoid robots. The topics related to our final goal are as follows; elucidating slip phenomena in turning motion, creating a model that predicts the amount of slip, and the realization of twirl exploiting slip with a humanoid robot. In the field of ergonomics and biomechanics, intensive research has been performed on analyzing slipping and slip induced in a fall[1] [2] [3]. However, there still exists some controversy on the conditions which result in falling [4]. Conventional studies on biped locomotion in low friction environment mainly focus on the prediction, avoidance, or recovery from the slip. Boone and Hodgins[5][6] simulated a running biped on a floor with a low friction area, by introducing a reflex control strategy. Park and Kwon[7] designed a controller to enlarge the frictional force at slipping, and simulated a 12-DOF biped robot walking on slippery surface. Kajita[8] described a calculation of ZMP concerning slip. Though ZMP[9] is useful to check the possibility of falling down, the conventional way to calculate ZMP cannot be used in case of slip. It provides a good prediction of falling caused by slipping. They also successfully demonstrated a walk on low friction environment using the humanoid robot HRP-2. There exists some work that addresses the issues regarding the proactive use of slip for humanoid motion. Nishikawa[10][11] proposed a mechanical method for biped 978-1-4244-2822-9/08/$25.00 ©2008 IEEE robots to use slip by adding a piston to the heel of the foot to generate rotational slip. Although the proposed mechanism is based on the idea of generating friction torque, the physical model is not numerically revealed. There also exists a study on the application of slip to the turning motion of a small humanoid robot HOAP-2 by Koeda[12]. The robot pivots the foot keeping contact with the floor, instead of lifting the foot. They claim that the robot is able to keep large support polygon while turning. The proposed motion used point contact between one corner of the pivoting foot and the ground for 90-deg turning motion. Fig. 1 shows a photograph sequence of a professional walking model performing a 360 degrees turn. By slipping the soles of both feet, she turns speedily and smoothly. Aiming our sights on this motion, we start investigation into the fundamental properties of rotational slip during a twirl. Fig. 1. Rei) 360 degrees turn by a professional walking model (Walking Studio To realize desired twirling motion, the relation between a robot motion and the resulting angle is investigated in this paper. We want to achieve a better understanding of the phenomenon of rotational slip through this study. The rest of this paper is organized as follows. After presenting the fact that such slip motion is used by humans based on actual human motion capture data, we make a hypothesis that the generated turning motion is caused by the effect of minimizing the power generated by floor friction in Section III. According to the hypothesis, we examine pre-calculated pattern and verify the adequacy of the proposed model in Section IV. In Section V, 90-deg ”twirl” utilizing both feet slip is successfully demonstrated by humanoid robot HRP-2. We conclude this paper and address our future plans in Section VI. II. H UMAN T URNING M OTION A NALYSIS We captured various turning motion performed by a professional walking model using a 3D optical motion capture system. Fig. 2 is a plot of foot placement during the 180 degrees turn made by the walking model. Fig. 3 shows the direction of the body (waist) of the model, and relative angles of both feet. The area with background color signifies the slipping phase. A twirl procedure obtained from analysis of the captured motion is described as follows: a: Landing on the left foot twisting toward the direction of rotation (Second step in Fig. 2). The body and the right leg start turning left. b: Landing on the right foot (Third step in Fig. 2) twisting and sliding toward the direction of rotation. The body still remains at a shallow angle. c: Starting slip of the left foot. d: Stopping rotation of the right foot. The body and the left foot are still moving. e: Stopping rotation of the left foot and the body. It restarts motion to the next step (Fourth step in Fig. 2). Both feet rotate during the period between c and d in the figures. Compared with a 90, or 360 degrees turn, it was observed that the larger the total angle of rotation becomes, the larger the angle in this period is. On the other hand, rotational slip does not occur during b to c when the total angle of rotation is small. The step just before the turn, the second step in Fig. 2, also tends to twist in correlation to the total angle of rotation. The data set was not large enough to supply the relations with quantitative evidence, thus we can only say that they have such tendencies. Human turning motions are realized not only statically through the trajectory of the feet but also through dynamics such as the moment of inertia caused by the twisting upper body or a swinging arm. However we do not go over the dynamical effects in this paper. In the rest of this paper, we focus on the rotational slip of the feet and the twirl of the body caused by it. Fig. 2. Foot placement during 180[deg] turn by a professional walking model. The red polygons represent the footsteps of right foot, and blue represents the left. III. ROTATION M ODEL FOR E XPLOITING S LIP Fig. 3. Waist pose and relative foot angles during 180[deg] turn by a professional walking model. The red area represents the slipping phase of the right foot, and blue represents the left. Purple represents simultaneous slip with the both feet. In this section, a model of rotation caused by friction force is proposed. In this proposition, we use the motion as shown in Fig. 5. The left hand side of Fig. 5 shows the given motion of feet. The dotted boxes represent the positions of the feet at the beginning of the motion. The robot stands with its feet 2d apart along the X axis, and 2b apart along the Y axis. The robot moves them according to the arrows and reaches the positions represented with solid boxes. The given trajectories of the feet are both parallel to the x-axis with a distance of b. Our assumption is that the robots body will rotate from the reaction force. The expected final positions of the feet are shown in the right hand side of Fig. 5. In the figure, the coordinate axis xW yW signifies the world coordinate, and xB yB signifies the floor projection of the base frame of the robot which origin is on the pelvis as shown in 4. At the start of the motion, the two coordinates are exactly the same. We define theta to be the angle between the world coordinates and the robot base coordinate. When the robot moves, we expect theta to increase. We make a hypothesis that the generated turning motion is caused by the effect of minimizing the power generated by floor friction. To simplify, we adapt the following assumption used by Harada[13] which is based on the method proposed by Lynch[14]: 1) The model assumed quasi-static motion, in other words, the motions are slow enough that frictional forces dominate inertial forces. 2) The dynamic and the static coefficient of friction acting between the feet and the floor are the same. 3) The plane of floor is uniform and horizontal, and the friction distribution on the sole of the foot does not change. According to the assumption 1, we deal in the twirl only by mechanistic effect of the lower body motion, and the turn built on the momentum of whole body will be covered in future work. In this paper we also assume the robot applies the same weight to the right and left feet, uniformly. Fig. 6 shows the motion of the right foot for explanation of the hypothesis. The point P is on the sole of a foot, and it slides to P while the robot moves its feet. The point P and P are at a distance of (x, y) from the center of the sole in the longitudinal and lateral direction, and their positions are expressed as follows: Px + x (1) P = Py + y Cθ −Sθ x P = (2) Sθ Cθ Py + y Fig. 4. Model of a robot and its base frame, and world frame. where, Px and Py are the position of the center of the sole in the world frame at the beginning, and Sθ = sinθ, Cθ = cosθ. The loss of the power at P can be calculated by multiplying the frictional force applied to small area at P by the sliding distance |P − P |: WP f |P − P | = (3) where f is the frictional force per unit area. Integrating over the area of sole, we obtain expressions for the total loss of the power at the sole of right foot: Wright = Fig. 5. Motion of both feet: The left shows the given pattern, the right shows the expected motion. The rotational torque may occur by friction force. = ly 2 ly 2 ly 2 lx 2 WP dxdy − l2x − l − 2y lx 2 f |P − P | dxdy − l2x (4) where lx and ly are the length and width of the sole, respectively. The frictional force per unit area is given by the coefficient of friction multiplied by the normal force: f μM g 1 2 lx ly = (5) where μ, M , and g are the friction coefficient, the total mass of the robot, and the gravity coefficient, respectively. In practice, since the applied normal force f is equal at any point of the sole, we obtain: Wright Fig. 6. Hypothesis to explain the slip phenomenon = μM g 2 ly 2 l − 2y lx 2 − l2x |P − P | dxdy (6) The loss of the power on the left sole is the same as the right since its foot motion is symmetrical to the right about the origin of the base frame of the robot. Substituting Eq. 1 and 2 into Ew. 6, we obtain the total loss of the power at the both soles W: W = μM g |P − P | − ly 2 lx 2 |P − P | dxdy − l2x (7) 2 T Cθ x Px + x − −Sθ Py + y 2 Sθ x Py + y − (8) Cθ Py + y 2 ly 2 = + In order to calculate θ that would minimize W, we do the following. Differentiating Eq. 7 with respect to θ yields: ∂W ∂θ ∂ μM g ∂θ = ly 2 − ly 2 lx 2 − l2x |P − P | dxdy (9) Since θ that would minimize W requires Eq. 9 to be zero, we obtain: l2y l2x ∂ |P − P | dxdy = 0 (10) ∂θ − l2y − l2x the proposed model. In these simulation and experiment, we keep to the same motion as shown in the last section. In the experiment, the robot moves the feet back and forth using various distances of d[m]. The start position of the right foot varies over d = {−0.08, −0.06, −0.04, −0.02, 0.02, 0.04, 0.06, 0.08, 0.10}[m]. The sideway distance of the feet are kept at 2b = 0.19[m]. The vertical axis rotation of the body at the final position is measured. For the simulation, we used OpenHRP, which is a dynamic simulator developed in the Humanoid Robotics Project(HRP)[15]. Simulated robot model is HRP-2, a 30 DOF humanoid robot which was also developed in HRP. Fig. 8 illustrates the simulation result at μ = {0.1, 0.2, 0.3, 0.4, 0.5}. In the simulation we found that the friction between feet and floor do not affect the slip angle. This result, that the friction coefficient does not affect the rotation, agrees to our hypothesis proposed in the last section. Since this equation has neither friction coefficient, nor the total mass of the robot, or the gravity acceleration, the shape of the sole and given motion pattern alone decide the rotation of the robot. This also means that the friction coefficient does not affect the rotation. Eq. 10 can also be described by the square of |P − P | as follows: W∗ = ly 2 ly 2 ly 2 − = − ly 2 lx 2 − l2x lx 2 − l2x Fig. 7. Experiment: foot motion of HRP-2 slipping and turning on the low friction floor |P − P | dxdy 2 2 {(1 − Cθ )x + Sθ (y + Py ) + Px } 2 = + {(1 − Cθ )(y + Py ) − Sθ x} dxdy 4 (1 − Cθ ) 2lx3 ly + lx (ly + Py )3 − (−ly + Py )3 3 +2Sθ Px lx (ly + Py )2 − (−ly + Py )2 +4Px2 lx ly ∂W ∗ ∂θ = 4 3 Sθ lx ly + lx (ly + Py )3 − (−ly + Py )3 3 +8Cθ Px lx Py ly = 0 (11) Finally we obtain: tanθ = − 3Px Py 2 lx + ly2 + 3Py2 (12) IV. S IMULATION BY O PEN HRP AND E XPERIMENT WITH HRP-2 This section reports the results of a simple slip test with a humanoid robot by simulation and experiment in order to find out features of rotational slip, and to verify the adequacy of Fig. 8. Comparison between different floor friction coefficients. The position d of the right foot at start position on the x-axis and the rotation at the final position on the y-axis. The humanoid robot HRP-2 was used for the experiment. To obtain low friction, small plastic plates are put on both side of the foot of HRP-2, where contact with a floor, as shown in Fig. 9, and a large sheet was set on the floor. The friction coefficient between the two plastic sheets was 0.23. These sheets were used because the soles of HRP-2 are covered with a rubber-like material. The friction coefficient of the rubber soles were μ > 1.0 on the lab floor, which may cause excessive motor load when the feet rub against the floor. Thus Eq. 12 is changed as follows: Fig. 9. The left shows the feet of HRP-2 with plastic plates. The right shows the figure of the foot. In case of HRP-2, lx = 0.23[m], ly = 0.135[m], lyc = 0.04[m]. tanθ = − 3Px Py lx2 + ly2 + ly lyc + ly2c + 3Py2 (13) Fig. 10 shows a sequence of photographs of the feet of HRP-2. Fig. 11 illustrates the results comparing between the numerical solution of Eq. 13, the simulation and the experiment. They agree well with each other. It can be said that the proposed hypothesis is adequate, and that the simulation agrees rather well with experiment, so the result also shows accuracy of the OpenHRP simulator. Linear approximation is also illustrated in Fig. 11. The line for experiment with HRP-2 is not on the origin of the coordinate. The distribution of mass on both feet of the robot is not completely even. This may have displaced the robot’s center of mass during the experiment. Fig. 11. Comparison with numerical result of the proposed model, simulation, and real experiment . The position d of the right foot at start position on the x-axis and the rotation at the final position on the y-axis. are same as the experiment in section above. The pattern of the motion was designed in reference to a series of human turning motion, which has been analysed in Section II. The trajectory of the feet during the turn is shown in Fig. 12. The robot walks straight ahead before and after the 90-deg turn.The feet are moved in circles so as to fix the stance during the turn, and it is different from the linear motion used in the last section. We adopted straight feet path in the last section in order to deal in the simplest model. The reason for this change was because we worried that moving the foot twisted inwards towards the other foot, would cause a collision. We also expected the motion to look more natural if the distance between the feet was fixed. Fig. 12. Motion pattern of the feet. The left shows the second step of Fig. 2. The right shows the twirl with both feet slipping. Fig. 10. Experiment: foot motion of HRP-2 slipping and turning on the low friction floor A sequence of photographs of the demonstration of HRP-2 is shown in Fig. 13, and you can see that a smooth 90-deg turn has been realized. The 90-deg turn took around 2.5 seconds to complete. By our conventional way of pattern generation, the 90-deg turn motion normally spends over 8 seconds with many small steps. V. D EMONSTRATION OF T WIRL BY HRP-2 VI. C ONCLUSIONS This section demonstrates a 90-deg twirl exploiting slip using HRP-2. The conditions and parameters of the experiment A preliminary discussion on twirling a humanoid robot by utilizing the slip between the feet and the ground was Fig. 13. Demonstration of HRP-2 turn described. By using human motion capture data, we figured a procedure for the twirl by extracting the essence from human motion. We also conducted both simulation and experiment with the humanoid robot HRP-2. To generate the slip motion, we predict the amount of slip under the hypothesis that the turning motion is caused by the effect of minimizing the power generated by floor friction. Even though the calculated trajectory was similar to the experiment and the simulation, the difference warrants further investigation. A turning motion utilizing both feet slip is successfully demonstrated with the humanoid robot HRP-2. Because of the fewer DOF and limited joint range of HRP-2, it is impossible to stand on the toes to get smaller friction radius, or to twist its leg as far as a human. Thus the generated motion was far from a 360 degrees turn shown by the human. Furthermore, humans utilize inertial force by twisting the upper body and swinging the arms. Future work will also address this effect to achieve highly sophisticated human like motion. R EFERENCES [1] L. Strandberg, “On accident analysis and slip-resistance measurement,” Ergonomics, vol. 26, no. 1, pp. 11–32, 1983. [2] M. Redfern and B. Bidanda, “Slip resistance of the shoe-floor interface under biomechanically-relevant conditions,” Ergonomics, vol. 37, no. 3, pp. 511–524, 1994. [3] J. Hanson, M. Redfern, and M. Mazumdar, “Predicting slips and falls considering required and available friction,” Ergonomics, vol. 42, no. 12, pp. 1619–1633, 1999. [4] R. Brady, M. Pavol, T. Owings, and M. Grabiner, “Foot displacement but not velocity predicts the outcome of a slip induced in young subjects while walking,” Journal of Biomechanics, vol. 33, pp. 803–808, 2000. [5] G. Boone and J. Hodgins, “Reflexive responses to slipping in bipedal running robots,” in Proc. on IEEE Int. Conf. Intelligent Robots and Systems, pp. 158–164. [6] ——, “Slipping and tripping reflexed for biped robots,” Autonomous robots, vol. 4, 1997. [7] J. Park and O. Kwon, “Reflex control of biped robot locomotion on a slippery surface,” in Proc. on the 2001 ICRA, 2001, pp. 4134–4139. [8] S. Kajita, K. Kaneko, K. Harada, F. Kanehiro, K. Fujiwara, and H. Hirukawa, “Biped walking on a low friction floor,” in Proc. on IEEE Int. Conf. Intelligent Robots and Systems, 2004, pp. 3546–3552. [9] M.Vukobratović and J.Stepanenko, “On the stability of anthropomorphic systems,” Mathematical Biosciences, vol. 15, pp. 1–37, 1972. [10] M. Nishikawa, Japanese Patent P2005-238 407A, Sept. 8, 2005. [11] ——, Japanese Patent P2007-326 169A, Dec. 20, 2007. [12] M. Koeda, T. Yoshikawa, and T. Ito, “Stability improvement by slipbased turning motion of humanoid robot,” in The 25th Annual Conf. of the RSJ, no. 3H15, 2007, (in Japaqnese). [13] K. Harada, “Pushing manipulation for multiple objects,” Trans. of the ASME, vol. 128, pp. 422–427, 2006. [14] K. Lynch, “Estimating the friction parameters of pushed objects,” in Proc. on IEEE Int. Conf. Intelligent Robots and Systems, 1993, pp. 186– 193. [15] H. Hirukawa, F. Kanehiro, and K. Yokoi, “Hrp develops openhrp,” presented at the IEEE/RSJ IROS Workshop on exploration towards Humanoid Robot Applications, 2001.
© Copyright 2026 Paperzz