IROS2019国际学术会议论文集1965

Moving onto High Steps for a Four-limbed Robot with Torso Contact* T. Matsuzawa1, T. Matsubara2, K. Namura2, X. Sun2, A. Imai2, M. Ohkawara1, S. Kimura2, K. Kumagai2, K. Yamaguchi2, H. Naito2, T. Sato1, K. Terae2, M. Murakami1, S. Yoshida1, A. Takanishi3and K. Hashimoto4 AbstractIn this paper, we describe approaches to enable a four-limbed robot to get over a step higher than its leg with torso contact. The higher the step becomes, the more diffi cult it is for legged robots to get over from the viewpoint of stability and kinematic reachability. Torso landing contributes to improving stability and robustness of motion for moving onto high step because of lower center of mass (CoM) and larger support polygon, which is seldomly utilized by previous human-sized legged robots. The approaches in this paper consist of the following two components. As for hardware, spikes are arranged on the bottom of robot?s body for stable torso landing on a high step. As for motion generation, Sequential Quadratic Programming (SQP) is utilized to generate motion with torso landing to guarantee stability of robots during getting over the high step. From experiments, it is confi rmed that the four- limbed robot WAREC-1 succeeded in moving onto a step with the height of 865mm. I. INTRODUCTION Legged robots have a feature that its discontinuous footing can be moved, and many studies about walking on rough terrain have been proposed. In particular, as represented by humanoid robots, a system having the same numbers of degrees of freedom as humans has high versatility and is expected to be applied to various applications. One of these applications is the operation of the robots at disaster sites. Since the competition “DARPA Robotics Challenge Fi- nals” 1, many robots for disaster response have been developed. In the competition, it is required to conduct both previously known tasks and unexpected tasks. To deal with various tasks, most of the robots participated in the *This study was conducted with the support of Research Institute for Science and Engineering, Waseda University; Future Robotics Organization, Waseda University, and as a part of the humanoid project at the Humanoid Robotics Institute, Waseda University. This research was funded by ImPACT TRC Program of Council for Science, Technology and Innovation (Cabinet Offi ce, Government of Japan). It was supported by Grant-in-Aid for JSPS Fellows Number 18J13950. It was also supported by early-bird program, Waseda University. It was also partially supported by SolidWorks Japan K. K; DYDEN Corporation; and KITO Corporation whom we thank for their fi nancial and technical support. 1T. Matsuzawa, M. Ohkawara, T. Sato, M. Murakami and S. Yoshida are with the Graduate School of Advanced Science and Engineering, Waseda University, 41-304, 17 Kikui-cho, Shinjuku-ku, Tokyo 162-0044, JAPAN matsuzawa-takane.waseda.jp 2T. Matsubara, K. Namura, X. Sun, A. Imai, S. Kimura, K. Kumagai, K. Yamaguchi, H. Naito and K. Terae are with the Graduate School of Creative Science and Engineering, Waseda University. 3A. TakanishiiswiththeDepartmentofModernMechan- icalEngineering,WasedaUniversityandisthedirectorof theHumanoidRoboticsInstitute(HRI),WasedaUniversity. contacttakanishi.mech.waseda.ac.jp 4K. Hashimoto is with the School of Science and Technology, Meiji University and is a researcher at the Humanoid Robotics Institute (HRI), Waseda University.hashimotomeiji.ac.jp competition are legged type 2-4. After the competition, some disaster response robots have also been developed 5. At disaster sites, it is necessary to overcome various rough terrains such as rubble road, stairs and high steps. In order to make legged robots walk stably in these environments, stability control is required so that legged robots will not fall or turn over against external disturbances such as collapse of the road surface. For stabilization, it is desirable to move the leg fast to adapt to a sudden disturbance. For instance, Bigdog 6 developed by Boston Dynamics realized a stability control so that the robot did not fall over when the road surface deforms by moving its legs quickly. In addition, the humanoid robot ATLAS 7 can demonstrate high speed and high instantaneous force by utilizing hydraulic driving system and move onto a step with jumping motion. When it comes to electrically driven robots, however, it is diffi cult to perform high instantaneous force as the scale of legged robots increase by utilizing electric motors commercially available in terms of their mass, torque and speed. In some cases, jumping motion is realized in a system using electric motors such as ASIMO 8, but realization of a system capable of acquiring a force enough to jump as higher as the length of legs is still diffi cult. When legged robots get over the step in the same way of conventional walking motion, it becomes diffi cult to walk as the height of step increases. When “the height of the step divided by the length of leg” is defi ned as step-leg ratio, the movable range of feet required at the time of getting over the step becomes large as the step-leg ratio increases. Especially when step-leg ratio exceeds 1, it is diffi cult to moving onto steps while securing that projection of the Center of Mass is within the support polygon. the ATLAS 7 mentioned above can overcome a step with jumping motion, but it is not confi rmed that the robot can overcome a step higher than its legs. Therefore, we propose a method for moving on rough terrain utilizing the contact of body and ground. This method consists of the development of a spike mechanism for improving the capability of moving on rough terrain and the strategy for overcoming rough terrain. In the proposed method, legged robot contacts its feet and body with rough terrain. That is, utilizing its body as feet. In addition, the body can also make a support polygon since the robot can be treated as a fi ve-legged robot. The novel point of the proposed method is to utilize torso landing of legged robots positively, which contributes to improving stability and robustness of motion for moving onto a high step. The proposed method is seldomly utilized by 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) Macau, China, November 4-8, 2019 978-1-7281-4003-2/19/$31.00 2019 IEEE6324 previous human-sized legged robots. In addition, the method makes it possible for legged robots to overcome a step whose step-leg ratio exceeds 1 even if the robot does not have a high instantaneous power. The key point of the spike mechanism is that it is designed to create a catch on rough terrain and can make a support polygon by hooking the spike to the edge of stairs or steps whose step-leg ratio exceeds 1. In addition, it also helps reducing the sliding distance of the robot by getting caught by the surface of rough terrain. Furthermore, since the Center of Gravity of the robot is lowered when the body is in contact with ground, there is the advantage that the impact force at the time of falling can be reduced. In this way, the method of utilizing the catch is useful in that it can improve the capability of moving on rough terrain even if the robot can not demonstrate high instantaneous power. Several studies about the movement utilizing the contact of the body and ground such as LittleDog 9, ANYmal 10, etc. have been conducted. The important point for legged robots moving on rough terrain is how to generate the grip force at the moment of grounding. Several researches ensuring grip force on rough terrain are as follows. Hirose et al. developed a mechanism which deforms with the surface of rough terrain by utilizing a bag fi lled with powder for contacting rough terrain 11. Parness et al. developed the four-limbed robot LEMUR-3 with small needle spikes to create catch and realized vertical wall climbing 12. For crawler robots like Quince 13, a large number of protrusions are arranged on the surface of the crawler belt. As the crawler rotates, the number of times the protrusion contacts rough terrain increases during traveling, and it is possible to create catch when moving on rubble road or stairs. Hutter et al. developed the four-legged robot ANYmal 10 which can go up and down stairs by utilizing protrusions mounted on the bottom of its body. The authors have developed four-limbed robots 1415, realizing the movement on rubble road by crawling motion that the robot goes by grounding its torso and feet alternately. In addition, a linear-spike mechanism is mounted on the body to reduces the sliding distance during moving on rough terrain 16. The rest of this paper consists of the following contents. In chapter 2, we explain the required specifi cation of the mechanical design, the method of determining the mecha- nism, and the validity verifi cation method for the developed spike mechanism mounted on legged robots. In chapter 3, we explain the strategy for getting over a high step. In chapter 4, the developed spike mechanism is mounted on a four-limbed robot, and the robot moves on rubble, stair and high step to confi rm the effectiveness of the proposed spike mechanism. Finally, in chapter 5 we discuss conclusions and future works. II. HARDWARE A. Requirements As is mentioned in the previous chapter, it is effective to hook the body to the road surface when using the ground contact of legged robots to overcome rubble road, stairs and Fig. 1.Overview of WAREC-1 and targeted high steps steps. Fig. 1 shows the overview of the four-limbed robot WAREC-1 15. The height of the robot is 1690 mm, and the mass is 150 kg. In addition, this robot consists of 28 degrees of freedom, and each limb has 7 degrees of freedom. When the robot contacts its torso on stairs, the robot tends to fall over as the angle of inclination increases. To make the conditions for climbing stairs strict to the biggest degree, we set the stair 17 whose inclination is approximately 45 as the target, which has the largest angle determined by the Building Standards Law of Japan. With regard to steps, it becomes diffi cult to get over the step as the step-leg ratio increases. Therefore, we set the step whose step-leg ratio is more than 1 as the target shown in Fig. 1. In other words, the height of the steps would be more than 800 mm in the case of WAREC-1. B. Design of Spike Mechanism It is required to satisfy the following conditions for de- signing landing mechanism of torso: can make support polygon as large as possible can create grip force on rough terrain by hooking can land on stairs without falling down can withstand a strength for force caused by self-weight In order to satisfy these requirements, multiple spikes are placed on the bottom of torso. The size of the spike should be determined on the condition that the robot can make the support polygon without falling over when the body is grounded to the stair with the inclination of 45. Also, to increase the normalized energy stability margin (NESM) 18 when the torso is landed, it is needed to decide the arrangement of spikes. Furthermore, to increase the catching on rough terrain, it is necessary to consider whether (i) the method of increasing the number of spikes and (ii) the method of designing so that the spike moves in the longitudinal direction are effective. To determine the design parameters of the spike, we designed four kinds of models and conduct an evaluation experiment. Model A: All spikes can be moved passively. Model B: Inside spikes can be moved and outside spikes are fi xed. Model C: All spikes are fi xed with two types of length. Outer spikes are longer than inner spikes. Model D: All spikes are fi xed with three types of length. Outer spikes are longer than inner spikes. 6325 Fig. 2.Mock-up model with spikes Fig. 3.Cross section of mock-up spikes As for using dynamics simulator, it is diffi cult to verify which body model is effective because it is diffi cult to accurately reproduce the elastic coeffi cient and friction of the passive mechanism and the friction between spikes and the road surface. Therefore, we actually made four types of mock-up bodies as shown in Fig. 2 and conduct tests with rubble road and stairs. To make spikes of WAREC- 1s torso land on stairs with inclination of 45designed by Japanese Industrial Standards (JIS) and United States Military Standard (MIL-SPEC) , the minimum length of a spike is calculated to be 25 mm. In addition, to make the torso shape like Model D, spikes with the length of 10mm and 40mm are added. More details of design for linear spike with spring and fi xed spike are shown in Fig. 3. C. Spike Arrangement To determine the arrangement of spikes, evaluation ex- periments about the number of falling over and sliding distance on three types of road surface shown in Fig. 4 were conducted. In order to unify the condition of the road surface, we use the random step 19 defi ned by NIST (National Institute of Standard and Technology). In the test on rubble road, we drop the mock-up body with a crane on top of the random step and measured the sliding distance from the point where the body grounded to the point where the body stopped by motion capture device (OptiTrack V120: Trio) in the horizontal direction. As for tests on stairs, we drop the mock-up body onto the stairs with a crane and calculated the success rate of stuck catches. The mass of the body is set to be 150 kg, which is approximately the same as the weight of WAREC-1. The number of trials is 5 times. Fig. 4. Random step defi ned by NIST and stairs Fig. 5.Experiment of mock-up body on random step TABLE I EXPERIMENTAL RESULT OF MOCK-UP BODIES HillDiagonalStair FallSlipFallSlipFall Model A1725-5 Model B0765-4 Model C0875-4 Model D0435322 The state of the experiment using the random step is shown in Fig. 5. This result shows that Model D is the most effective one. However, as shown in Table I, the average sliding distance on rubble road is 43 mm, which is 21.5% of 200 mm, which is the distance per crawling motion of WAREC-1. Furthermore, the number of falling occurred on 2 kinds of random step was 2 in 5 trials (40%). From the result mentioned above, it was confi rmed that the body could not be expected to improve the moving speed on rough terrain and there was the risk of falling. The reason why the sliding distance became larger was that the number of spikes was small. In addition, the cause of the falling down of the body on stairs is that the supporting polygon is small and NE stability margin could not be suffi ciently guaranteed. Therefore, we developed a new mock-up body with spikes added to the position near the base of limbs as shown in Fig. 6. The interval of spikes was set to be 76 mm and the maximum length of spike was set to be 60 mm based on the following conditions: (i) to make spikes be able to be landed on stairs designed based on Japanese Industrial Standards (JIS) and United States Military Standard (MIL- SPEC) and (ii) to make spikes be arranged at equal intervals and as far as possible from the center of body surface to expand support polygon. As the support area expanded, the number of spikes increased to 34. Experimental results using the new mock-up body are shown in Table II. The success rate of falling on each random step increased from 40% to 100%. The number of falling on 6326 Fig. 6. Design of refi ned mock-up body with spikes TABLE II EXPERIMENTAL RESULT OF THE REFINED MOCK-UP BODY HillDiagonalStair FallSlipFallSlipFall Refi ned model060100 each random step was 0, and the sliding distance was reduced to 10 mm. This sliding distance corresponds to 5% of 200 mm. As a result, experiments mentioned above confi rmed that the following contents are effective for preventing the falling and slippage when the torso is landed on the rubble road and stairs. Making the length of spikes fi xed Arrangement that spikes at each corner are long and spikes become shorter as going to the center of the body Placing some spikes on the outside of the body for expanding support polygon D. Details of Spike Mechanism Fig. 7 shows the developed body based on the experiment described in the previous section, which has 34 cylindrical spikes as shown in Fig. 8. The length of spikes is the longest at four corners of 60 mm and becomes short as 50 mm, 40 mm, 30 mm, 20 mm as going to the center, and the torso surface is concave as shown in Fig. 1. The interval of spikes is set to 76 mm so that spikes can be caught in stairs. The shape of spikes made by A7075-T6 is designed not to cause plastic deformation even with a shear load of 1500 N, which is its own weight of WAREC-1. In addition, spikes are designed to enable a factor of safety to exceed 3.0 considering the impact force with falling down and slippage. Moreover, in order to reduce sliding distance of the body when grounded on rough terrain, urethane rubber for improving the frictional force with the road surface is placed on the surface of the body and spike tip. III. MOTION OF STEP CLIMBING A. Motion Strategy An important point in getting over the step is to move the projected point of the Center of Gravity of the robot onto the road surface above the step as shown in Fig.9. It is Fig. 7.Developed spike mechanism mounted on WAREC-1 Fig. 8.Overview of a developed spike expected that debris of extreme environments is scattered in various shapes and sizes of rubble. There are countless types of steps from simple to complex shapes, so it is required to consider various step differences in the strategy of getting over steps. If the robot can move the projected point of the Center of Gravity above the step, the robot can overcome the step by performing an action that falls over the step after that. The strategy of “falling onto the step after moving the projected point of the Center of Gravity of the robot on the road surface to the top of the step” is very simple, and can be applied to vari