Integration of Manipulation and Locomotion by a Humanoid Robot

Transcription

1 Integration of Manipulation and Locomotion by a Humanoid Robot Kensuke Harada, Shuuji Kajita, Hajime Saito, Fumio Kanehiro, and Hirohisa Hirukawa Humanoid Research Group, Intelligent Systems Institute National Institute of Advanced Industrial Science and Technology(AIST), JAPAN. {kensuke.harada, s.kajita, f-kanehiro, hiro.hirukawa}@aist.go.jp Abstract. We aim to develop a humanoid robot which can go anywhere an ordinary human covers. For such purpose, we consider enhancing the walk and work capabilities of a humanoid robot by integrating manipulation and locomotion. Especially, we focus on three motions of a humanoid robot: pushing a heavy object placed on the floor, working while thrusting a hand on to a desk, and ascending large steps while holding a handrail for support. These motions are realized by experiment using the humanoid robot HRP-2. 1 Introduction In recent years, several humanoid robots[1 3] have developed. They can walk on the flat floor, rough terrain with small gaps, and climb up stairs. On the other hand, a humanoid robot is expected to work instead of a human in a real environment. Since such an environment is not structured for a robot, it often includes rough terrain with large gaps and large obstacles. However, by actively using arms for propping the body, we can expect that a humanoid robot can extend walking and working capabilities. For enhancing the walking and working capabilities, a humanoid robot can use two, three or four hands/legs as occasion requires. While we can consider many styles of integration of manipulation and locomotion, we focus on three styles of them: (i)pushing manipulation of a large and heavy object placed on the floor, (ii)handling task while supporting the body with an arm propping against a table, and (iii)climbing up a large gap while holding a handrail for support. These motions are considered to be essential when a humanoid robot works in a real environment. When a hand of a humanoid robot contacts an environment, one of the difficulties is in the theoretical analysis of the robot s balance. Conventionally, the balance control of bipedal walk in humanoid robot has been mostly ensured based on a reference index called ZMP, while the ZMP requires all the contact points lying on a plane. On the other hand, once the hands of a humanoid robot contacts an environment, it becomes difficult to judge whether or not the robot can keep balance. For such problem, we have been extended the ZMP analysis to manipulation tasks[7,9]. Also, when a humanoid robot manipulates an object, the reaction force is applied at the hands. If the reaction force is large, it becomes difficult for a humanoid robot

2 2 K. Harada, S. Kajita, H. Saito, F. Kanehiro, and H. Hirukawa to keep balance. For such problem, we have been studied the force control based manipulation method[8]. On the other hand, this research studies the experiments on integration of manipulation and locomotion by a humanoid robot. We will focus on the practical aspects for realizing such integration. In this paper, after showing the previous works in Section 2, we show the experimental setup imitating a construction site in Section 3. The experimental results are shown in Sections 4, 5, and 6. 2 Related Works Recently, several humanoid robots such as ASIMO[1], QRIO[2], johnnie[3], H7[4], and HRP-2[5]. As for the research on manipulation by a humanoid robot, Yokoyama et al.[10] realized by the humanoid robot HRP-2 carrying a panel with a human. Inoue et al.[11] determined the foot position maximizing the manipulability of the arms. Takenaka[12], the authors[6,8], Hwang et al.[13] also studied pushing task by a humanoid robot. The authors[7,9] also studied the ZMP applicable for the manipulation tasks. 3 Problem Statement Fig.1 shows the overview of the experimental environment. This experimental environment imitates a construction site. Pushing a Large Object: This experimental environment includes a large object. The weight of the object is 25.9[kg] which is about half of the robot s weight. The coefficient of static friction between the object and the floor is about In this research, a humanoid robot will push and move the object with keeping balance of it. Handling Task: As shown in Fig.1(a), there are structures made of steel pipes. These structures imitate the scaffolding which can be seen in a construction site. As shown in Fig.1(c), the structure includes a stage whose height is 50[cm]. Manipulation of an object placed on a stage is sometimes difficult for a humanoid robot. This is because, if the object is far from the robot, the humanoid robot cannot capture the object while standing on two legs. For such problem, we consider making a humanoid robot manipulate the object while supporting the body with an arm propping against the stage(fig.2). In this experimental environment, the distance between the object and the robot is set to be 80[cm]. Climbing a Large Step: As shown in Fig.1(b), the experimental environment also includes a large step whose height is 28[cm] and a handrail beside the step. For a human-sized humanoid robot, it has been difficult to climb up such a large step. There are two reasons for the difficulty: (1)since a humanoid robot has to move largely during the single support phase, it becomes difficult to keep balance, and (2) it sometimes becomes difficult to generate the collision free motion within the movable range of each joint. As for the

3 Integration of Manipulation and Locomotion by a Humanoid Robot 3 Stage Scaffolding Pushed Object Big Step Handrail (a) Overview of Experimental Environment (b) Step, Handrail, and Pushed Object Fig. 1. Experimental Setup (c) Stage in the Scaffolding (d) The Humanoid Robot HRP-2 second reason, the humanoid robot H7[4] realized to climb a big step whose height is 25[cm] by utilizing the toe joint. In this research, a humanoid robot will climb up a large step whose height is 28[cm] while holding a handrail. By holding a handrail, we can expect that the robot would become more stable. As a humanoid robot, we use the HRP-2[5] developed in the humanoid robotics project[17]. The total DOF of HRP-2 is 30 (2DOF Head, 2DOF Waist, two 6DOF Arms, two 1DOF Hands, and two 6DOF Legs). And, the total weight of it is 58 [kg]. Since HRP-2 has 2DOF waist joint, various whole body motion can be generated. Some of the motions in this research are generated utilizing the waist joint. From the next section, we show the experimental results of (i)pushing a large object, (ii)handling task, and (iii)climbing a large step. 4 Pushing a Large Object Let the position of the ZMP, the COG(Center of Gravity) of the robot, and each hand be p zmp =[x zmp y zmp z zmp ] T, p G =[x G y G z G ] T,andp Hj =[x Hj y Hj z Hj ] T (j =1, 2), respectively. Let the hand reaction force and the total mass of the robot be f j =[f xj f yj f zj ] T (j =1, 2) and M, respectively. The relationship between the ZMP position and the hand reaction force is expressed by[6]: x zmp = L Gy + Mx G ( z G + g) M(z G z zmp )ẍ G M( z G + g)

4 4 K. Harada, S. Kajita, H. Saito, F. Kanehiro, and H. Hirukawa Fig. 2. Handling Task while Supporting Body with an Arm 2 j=1 (z Hj z zmp )f xj, (1) M( z G + g) y zmp = L Gx + My G (ż G + g) M(z G z zmp )ÿ G M( z G + g) 2 (z Hj z zmp )f yj, (2) M( z G + g) j=1 where L G =[L Gx L Gy L Gz ] T denotes the angular momentum of the robot about the COG. In eqs.(1) and (2), the first term of the right hand side shows the ZMP position without considering the hand reaction force, while the second term shows the displacement of the ZMP position due to the hand reaction force. From the second term of the right hand side of eqs.(1) and (2), we can see that the lower the position of the COG be, the smaller the effect of hand reaction force onto the ZMP position becomes. The robot can keep the dynamical balance if the ZMP is included in the foot supporting area. However, through the experiment, we observed that the robot tends to be unstable if the contact force is applied at the hands in the single support phase. Therefore, we consider pushing the object during the double support phase where both of the feet contact the floor and where the convex hull of the foot supporting area becomes wide. And, the robot steps without pushing an object[8]. The experimental result of pushing manipulation is shown in Fig.3. First, a humanoid robot steps without pushing an object as shown in Fig.3(b). To reduce the effect of hand reaction force on the ZMP position, the humanoid robot squats down as shown in Fig.3(c). As shown in Fig.3(e) and (f), the humanoid robot pushes an object where the amount of displacement of the object is 20[cm]. Here, to compensate the error caused in the ZMP position due to the hand reaction force, the position of the feet is modified based on the second term of the right hand side of eqs.(1) and (2). After finished pushing an object, the robot steps back as shown in Fig.3(h) and (i). In this experiment, while the humanoid robot pushes the object only once, the

5 Integration of Manipulation and Locomotion by a Humanoid Robot 5 experiment of pushing manipulation with continuously walking can be seen in the paper[6,8]. (a) t=0.0[sec] (b) t=3.6 (c) t=7.2 (d) t=10.8 (e) t=14.4 (f) t=18.0 (g) t=21.6 (h) t=24.2 (i) t=27.8 Fig. 3. Experimental Result on Pushing Manipulation 5 Handling Task When manipulating an object placed on the table, it sometimes become difficult for a humanoid robot to keep balance while standing on two legs. Thus, we utilize the interaction between the hand and the table to manipulate the object. The overview of the proposed method is shown in Fig.4. This method is just similar to the crawl gait of a quadruped walking robot. First, both of the hands of a humanoid robot contacts the table. Then, the waist of the humanoid robot moves right making the horizontal projection of the COG be included in the convex hull of the horizontal projection of the supporting points in the right hand, right foot, and the left foot. Since the left hand can detach from the table, the humanoid robot moves to make the left hand be close to the object. If the humanoid robot cannot capture the object, then the humanoid robot moves to make the left hand contact the table. In such a case, the waist of the humanoid robot moves left making the horizontal projection of the COG be included in the convex hull of the horizontal projection of the supporting

6 6 K. Harada, S. Kajita, H. Saito, F. Kanehiro, and H. Hirukawa Target Object COG Feet Fig. 4. Overview of the Proposed Method (a) t=0.0[sec] (b) t=5.8 (c) t=11.7 (d) t=17.6 (e) t=23.4 (f) t=29.27 (g) t=35.1 (h) t=41.0 (i) t=42.0 Fig. 5. Experimental Result on Hnadling Task points in the left hand, right foot, and the left foot. Then, the humanoid robot tries to manipulate the object by the right hand. The result of experiment is shown in Fig.5. As shown in the figure, the humanoid robot succeeded in manipulating the lever while supporting the body with an arm. We note that, while we do not consider the dynamics of the robot, we can obtain the condition for the ZMP for keeping the dynamical balance of the robot by using our previous work[7]. 6 Climbing a Large Step When climbing a large step, a humanoid robot can increase the stability by utilizing the interaction between the hand and the environment. We can also expect that,

7 Integration of Manipulation and Locomotion by a Humanoid Robot 7 (a) t=0.0[sec] (b) t=3.75[sec] (c) t=7.5[sec] (d) t=11.25[sec] (e) t=15.0[sec] (f) t=18.75[sec] Fig. 6. Overview of the Planned Motion (g) t=22.5[sec] (h) t=26.25[sec] (i) t=30.0[sec] by grasping a handrail, it becomes easier for a robot to keep balance. The basic mechanism of ZMP for a humanoid robot grasping a handrail is shown in the appendix. We can see that, the stronger the humanoid robot grasps a handrail, the easier the humanoid robot can keep balance. The motion of a humanoid robot climbing a large step is shown in Fig.6. First, a humanoid robot grasps the handrail. Then, a humanoid robot puts the left foot onto the step. The humanoid robot further moves to make the position of COG forward as shown in Fig.6(e). For this posture, the humanoid robot cannot keep balance unless grasping the handrail. The robot finally climb up the large step as shown in Fig.6(i). The experimental results are shown in Fig.7. 7 Conclusion In this paper, for the purpose of developing a humanoid robot that can go anywhere an ordinary human goes, we have shown three basic motions, i.e., pushing a large object, handling task while supporting the body with an arm, and climb a large step while grasping a handrail. When performing the experiment, we assume that the geometry of the environment is known. In our future work, we will use the vision sensor when performing an experiment. This work has been carried out under the Basic Technology Research Promotion Program "R & D of Basic Technology for Humanoid Robot Working in Real

8 8 K. Harada, S. Kajita, H. Saito, F. Kanehiro, and H. Hirukawa (a) t=0.0[sec] (b) t=8.0 (c) t=16.0 (d) t=24.0 (e) t=32.0 (f) t=40.0 (g) t=48.0 Fig. 7. Experimental Result of Climbing a Large Step (h) t=54.0 Environment" of the NEDO, FY We would like to express our sincere gratitude for Dr. Kenji Kaneko, Dr. Kazuhito Yokoi, Mr. Kiyoshi Fujiwara who are the humanoid robotics researchers in AIST for their helpful discussions. We would also express our sincere gratitude for Mr. Masaru Nakamura with Kawada industries. He helped us a lot when constructing the experimental environment. References 1. M. Hirose, Y. Haikawa, T. Takenaka, and K. Hirai, Development of Humanoid Robot ASIMO, Workshop of IEEE/RSJInt. Conf. on ItelligentRobots and Systems, Workshop2, Y. Kuroki, M. Fujita, T. Ishida, K. Nagasaka, and J. Yamaguchi, A Small Biped Entertainment Robot Exploring Attractive Applications, Proc. of the 2002 IEEE International Conference on Robotics and Automation, pp , M. Gienger, K. Löffler, and F. Pheiffer, Towards the Design of a Biped Jogging Robot, Proc. of the 2001 IEEE International Conference on Robotics and Automation, pp , S. Kagami, K. Nishiwaki, J.J. Kuffner, Y. Kuniyoshi, M. Inaba, and H. Inoue: Online 3D Vision, Motion Planning and Bipedal Locomotion Control Coupling System of Humanoid Robot : H7, Proc. of IEEE Int. Conf. on Robotics and Automation, pp , 1998.

9 Integration of Manipulation and Locomotion by a Humanoid Robot 9 5. K. Kaneko et al.: Humanoid Robot HRP2, Proc. of IEEE Int. Conf. on Robotics and Automation, pp , K. Harada, S. Kajita, K. Kaneko, and H. Hirukawa: Pushing Manipulation by Humanoid considering Two-Kinds of ZMPs, Proc. of IEEE Int. Conf. on Robotics and Automation, pp , K. Harada, S. Kajita, K. Kaneko, and H. Hirukawa: ZMP Analysis for Arm/Leg Coordination, Proc. of IEEE/RSJ Int. Conf. on Intelligent Robots and Systems, pp , K. Harada, S. Kajita, F. Kanehiro, K.Fujiwara, K. Kaneko, K.Yokoi, and H. Hirukawa: Real-Time Planning of Humanoid Robot s Gait for Force Controlled Manipulation, Proc. of 2004 IEEE Int. Conf. on Robotics and Automation, pp , K. Harada, H. Hirukawa, F. Kanehiro, K.Fujiwara, K. Kaneko, S. Kajita, and M. Nakamura: Dynamical Balance of a Humanoid Robot Grasping an Environment, Submitted to 2004 IEEE/RSJ Int. Conf. on Intelligent Robots and Systems, K. Yokoyama, H. Handa, T. Isozumi, Y. Fukase, K. Kaneko, F. Kanehiro, Y. Kawai, F. Tomita, and H. Hirukawa: Cooperative Works by a Human and a Humanoid Robot, Proc. of IEEE Int. Conf. on Robotics and Automation, pp , K. Inoue, H. Yoshida, T. Arai, and Y. Mae: Mobile Manipulation of Humanoids Real-Time Control Based on Manipulability and Stability ", Proc. of IEEE Int. Conf. on Robotics and Automation, pp , Takenaka: Posture Control for a Legged Mobile Robot", Japanese Patent Application, H , Y. Hwang, A. Konno, and M. Uchiyama: Whole Body Cooperative Tasks and Static Stability Evaluations for a Humanoid Robot", Proc. of IEEE/RSJ Int. Conf. on Intelligent Robots and Systems, pp , F. Kanehiro et al.: Virtual humanoid robot platform to develop controllers of real humanoid robots without porting, Proc. of IEEE/RSJ Int. Conf. on Intelligent Robots and Systems, H. Hirukawa et al.: OpenHRP: Open Architecture Humanoid Robot Platform, Proc. of Int. Symp. on Robotics Research, S.Kajita et al.: Biped Walking Pattern Generation by using Preview Control of Zero- Moment Point, Proc. of IEEE Int. Conf. on Robotics and Automation, pp , H. Inoue et al.: HRP: Humanoid Robotics Project of MITI, Proc. of thefirstieee-ras Int. Conf. on Humanoid Robots, A ZMP considering the Grasping Force As shown in Fig.8, let us consider the situation where a hand of a humanoid robot grasps an handrail by modeling the robot by the cart-table model[16]. Here, for simplicity, we model the hand of a humanoid robot by a parallel gripper with a one-dof translational joint as shown in the figure. Let f H1 ( 0) and f H2 ( 0) be the reaction forces applied by the parallel gripper in the left-hand side and the right-hand side, respectively, of the handrail. Let f L1 ( 0) and f L2 ( 0) be the reaction forces between the table and the floor at the left-edge and the right-edge, respectively, of the contact segment, where f L1 = mg f L2 is satisfied. We neglect the effect of friction at each contact point

10 10 K. Harada, S. Kajita, H. Saito, F. Kanehiro, and H. Hirukawa τ ZMP 2 (a) A humanoid robot holding a hand-rail Fig. 8. Explanation of the proposed method (b) The cart-table model to explain the principle as simple as possible. Also, we consider the case where the acceleration of the cart is large making f H1 =0. As a source of ground reaction moment, we consider all forces acting on the cart-table system. Since the ground reaction moment τ Z around the ZMP is zero, it can be formulated as 0= f L1 x Z + f L2 (l x Z ) f H2 z H. (3) Solving eq.(3) with respect to the position of ZMP, the following equation can be obtained: x Z =(f L2 l f H2 z H )/(mg). (4) Now, let us focus on the region of ZMP for keeping the dynamical balance of the cart-table system. When both f L1 > 0 and f L2 > 0 are satisfied, the cart-table model keeps the dynamical balance. It is equivalent to 0 <f L2 <mg. Substituting this relationship into eq.(4), the set of ZMP keeping the dynamical balance can be obtained as: { } X Z = x z fh2 z H /(mg) <x z <l f H2 z H /(mg),f H2 0, (5) where we note that, since we only consider the case of high acceleration in this section, we focus on the lower limit of the set of ZMP. Thus the inequality x z < l f H2 z H /(mg) in the above equation does not make sense. This set of ZMP is a function of the grasping force of the gripper. If the grasping force is zero (f H2 =0), the set of ZMP is bounded by x Z > 0 which means that the ZMP exists inside of the contact area between the table and the floor. On the other hand, when the hand is grasping a handrail, the set of ZMP is bounded by x Z > f H2 z H /(mg) which means that the cart-table system can keep the dynamical balance even if the ZMP exists out-side of the contact area. It also means that, if the hand grasps an handrail with stronger grasping force, the robot can keep the dynamical balance more easily.