Proceedings of the 2011 IEEE International Conference on Robotics and Biomimetics December 7-11, 2011, Phuket, Thailand Design and Implementation of Humanoid Biped Walking Robot Mechanism towards Natural Walking Ren C. Luo and Yi Hao Pu Department of Electrical Engineering National Taiwan University Taipei, Taiwan { renluo & r98921008}@ntu.edu.tw Abstract This paper presents the implementation of the specifically designed humanoid biped walking robot legs towards natural walking. The humanoid biped walking robot legs employs ideas from both the active-actuated biped robot legs and the passive dynamic walkers. The approach is primarily based on the utilization of shock absorber and parallel linkage mechanism, which drastically decreases the requirement of actuator output force thus enables the humanoid biped walking robot legs to achieve higher efficiency while remain lower cost than traditional design. The specifically designed mechanism also provides better support than conventional humanoid biped walking robot. The walking algorithm that combines Series Elastic Actuation and Limit Cycle Walking permits the robot to move in more natural way. The experiment along with simulation and analysis demonstrate the high potential and possibilities of this concept and introducethe new direction towards the naturally walking humanoid robot legs design. Keywords- Humainoid Robot Legs; Series Elastic Actuation; Limit Cycle Walking; Combined Natural Active Walking and Passive Walking I. INTRODUCTION The Research of the humanoid biped robot has been developed for decades. Humanoid biped walking robots are capable of achieving multiple tasks, which conventional wheeled robots are unable to do, for example, walking on stairways or rugged terrain. Furthermore, humanoid biped robots mimic the motion behavior of human; therefore capable of provide quantitatively information (with some sensor mounted on the robot) about the effectiveness of the welfare and rehabilitation machine through proper experiment.for the purpose of making humanoid biped robot act resembling to human being, different approaches have been established. One common approach derives from the traditional robotics, model the humanoid biped walking robot leg as an inverted pendulum, deduce the overall transfer function and dynamic equation, and then applying the inverse kinematic to plan the trajectory of each joint to meet the ZMP criterion introduced by Miomir Vukobratovi. The major difficulty this approach may confront is that the overall transfer function and dynamic equation are too complicated to obtain. In addition, the output torque requirement of each joint is unimaginably high, thus enhances the trend of utilizing expensive harmonic drive and brushless motor, which in turn raises the threshold of entering the research field of humanoid biped robot. Chwan Hsen Chen, Jia Rong Chang and Cheng Yen Li Department of Mechanical Engineering Yuan Ze University Taoyuan, Taiwan mecch@saturn.yzu.edu.tw & {s975039 & s985042}@mail.yzu.edu.tw The humanoid biped robot Asimo developed by Honda is their twelveth-generationed product based on the paradigm described above [1][2][3]. It is capable of running and walking up and down the stairs, which are very impressive results. Wabian family from Waseda University is another example with the same design concept [4]. WABIAN-2R is capable of walking with stretched knee, heel-contact and toe-off motion [5]. WL-16 employs the mechanism similar to Stewart platform, driven by linear actuators and is able to carry one person. They are both cluster of Japanese researching highlight in this domain. In German, Researchers of Munich University of Technology also continuously devote themselves in humanoid biped robot area; Johnnie and its successor, Lola, are their very successive products. [6][7][8]. On the other hand, a completely different approach has been developed through the specifically designed mechanism of the humanoid biped robot legs. With the installation of compliant device such as spring and damper, the mechanism itself is capable of storing potential energy like muscles of human being and capable of walking along the ramp without any power source other than gravity. The research group from Delft University of Technology is the expert of this non-actuated passive humanoid biped walking robots. They have proposed many passive walkers during the past few years. Based on their original understanding in passive dynamic walker, they present Flame in 2008, which is an active-actuated humanoid walking robot [9]. The idea such as Series Elastic Actuation and Limit Cycle Walking are utilized in Flame and it successfully achieves a stable and human alike walking process [10]. However, the legs with only three activeactuated joints confine its ability and agility. Our goal is to exploit both the benefit of the active-actuated robots and the passive dynamic walkers, while active-actuated robots tend to have high versatility and thus are able to perform multiple tasks, and passive dynamic walkers have better performance in terms of efficiency and walk more human-like and natural. II. MAIN IDEA The main idea of our design originates from Flame. However, to improve the performance in terms of versatility and durability, some modifications are made. For the purpose 1165
of performing multiple tasks and mimic the human walking gait, our humanoid robot leg mechanism has five degree-offreedom for each leg as well as two passive joints whereas the Flame only has 3 degree-of-freedom and one passive joint. Fig. 1 shows the configuration of whole humanoid robot leg mechanism. Due to the same reason described above, the size and weight of the robot inevitably increases, which means every joint of the humanoid robot leg mechanism must sustain heavier load and thus is challenging the durability of the mechanism itself. The specific designed mechanism utilized the concept of vehicle suspension, where the linkage mechanism and the shock absorber are exploited in most of them. Combined with the bilateral linear actuators motion, we come up with an elegant solution that successfully takes advantage of best of the both worlds, well functions and meets every request of ours while only occupy a relatively limited space. III. MECHANISM DESIGN A. Series and parallel shock absorber Many humanoid robot legs suffer from damage after the long-term experiment since the shock from ground is too strong for the structure and motor shaft. To avoid such problem, the shock absorbers are utilized in our mechanism design. The shock absorbers employed in our humanoid robot legs are actually air springs used on mountain bikes. Fig. 2 is its picture. They are mounted on the humanoid robot legs in both series way and parallel way, as depicted in Fig. 3. Fig. 1. The entire configuration of the robot leg proposed. Fig. 2. The picture of air spring as shock absorber. Besides the effect as a buffer, the shock absorbers are capable of storing potential energy from the inertia of movement of the humanoid robot legs itself, which in turn leads to the lower requirement of the actuator output. B. parallel linear linkage mechanism The mechanism design of WL-16 from Wabian family inspires us to use the linear actuators combined with the parallel linkage mechanism in order to give better support to the humanoid robot. Conventional rotary DC motors used in humanoid robot legs are often heard suffering from lack of strength of its shaft, since the shaft not only has to transmit the torque but also has to defy the torque caused by gravity and ground shock. With our design, the torque caused by gravity and ground shock will be apportioned and spread through entire linkage mechanism structure. The humanoid robot LOLA developed in University of Technology Munich also employ a parallel drive in the ankle joint, the required motor peak torque is claimed to be reduced by approximately 35 % [11]. Most of the humanoid robots nowadays are using high power output brushless motor with harmonic drive to reach the incredibly high output torque requirement. With the adoption of parallel linkage mechanism and the parallel mounted shock absorber, the peak output force requirement of our linear actuator is enormously decreased, which enables us to choose linear actuator with lower power consumption while maintaining adequate output force and output speed. Fig. 4 shows the simple structure of the linear actuator chosen. Fig. 3. The basic diagram denotes the hybrid joint. 1166
absolute rotary encoder mounted at each joint. In force control mode, the commands sent to each local controller are reference output force of linear actuator, and the pressure sensors mounted on every shock absorber feedback the force. Nonetheless, the shock absorbers are not combined with every joint but only with the critical ones, which are the pitch of hip, knee, and ankle joints. The control of other joints is operated under position control mode. The Gyro sensor mounted on the waist of the robot will provide the information of overall orientation and enable the main controller to give complementary force command when the robot is losing its balance. Fig. 4. The structure of the linear actuator chosen. When two hip actuators move in identical direction, they will give sufficient force to drive the pitch motion in hip, whereas the different moving directions of two hip actuators will drive the roll motion in hip. When two knee actuators move in identical direction, they will give sufficient torque to drive the pitch motion in knee, whereas the different moving directions will drive yaw motion in hip. Ankle actuators are only able to move in common way and drive the pitch motion in ankle. With this special bilateral linear actuators motion, the requirement of the force output of actuators can be further decreased. Shock absorbers and linear actuators are also playing the critical roles of mimicking human muscle. According to human anatomy, our joints are propelled by the compress and stretch of muscle, which is similar to the linear motion of our actuators and shock absorbers. IV. MOTION CONTROLLER DESIGN A. Control Architecture Fig. 5 depicts the overall control architecture. The main controller is a windows-based PC. The command will be sent to local controller via can bus. Fig. 5. The overall control architecture. There are two kinds of the control loops. One is the position control, and the other is force control. In position control mode, the command sent by the main controller is the position of the linear actuator, with the feedback from the B. Limit Cycle Walking The position command from the main controller is generated with the newly introduced stability paradigm of Limit Cycle Walking, which is first developed by Delft University of Technology for their passive walkers with no actuator. The essence of Limit Cycle Walking is that it is possible to obtain a stable periodic walking motion sequence without verifying local stability at every instant of time during gait, which means it has fewer artificial constraints and thus easier to achieve a more efficient, natural, fast and robust walking motion. [12] By the latest work of Delft University of Technology, it has been shown that Limit Cycle Walking can also be implemented well on an active-actuated robot with multi degree of freedom. But Limit Cycle Walking still maintains many properties related to passive biped walker. Therefore the Limit Cycle Walker moves mostly according to the natural dynamic of the mechanism. In other words, it is very important to design the mechanism with high compliance. C. Series Elastic Actuation In order to meet the requirement, the actuators must have low impedance and high controllability. Combined the series shock absorbers in our design, the force control loop is actually a variation of the Series Elastic Actuation, which is the integration of the certain mechanism design and control architecture introduced by Jerry Pratt et al from Massachusetts Institute of Technology [13]. The Series Elastic Actuation consists of the linear actuator and the spring; they are mounted in series, so the force applied by actuator is the force on spring, with the sensor measuring the compression of the spring and feedback to the controller, the force applied by actuator can be easily calculated by Hooke s law of elasticity. The benefit of the Series Elastic Actuation includes lower output impedance and back-driveability as well as higher shock tolerance and the fidelity of force[13]. The difference between our design and conventional Series Elastic Actuation architecture is that the pressure sensors used to measure the force on air springs have to be mounted on both series and parallel shock absorbers to acquire the compression quantities of both shock absorbers, and then get the output force information Series Elastic Actuation is perfect for Limit Cycle Walking, because the natural dynamics of the walker itself is able to be intactly represented. 1167
A. Position Control Loop V. EXPERIMENTAL RESULTS Fig. 6. The simple diagram represent the relation As shown in Fig. 6, the command sent from local controller to the actuator is stroke (Denoted as S), however, the data acquire by rotary encoder is angle. The equation below demonstrate the relation of the stroke and angle, which is derived from the law of cosines. To establish a position control loop, the nonlinearity of the above relation is obviously undesired. Through the simulation in engineering software ADAMS, the simplified model of a single joint is built as shown in Fig. 7, the built-in postprocessor of ADAMS depicts the diagram as in Fig. 8, and clearly shows that in the particular joint angle and actuator stroke range of concern, the relation is rather linear,, which greatly reduce the complexity of control. B. Force Control Loop To acquire the accurate data of actuator output force, the force relation of the linear actuators, series shock absorbers and parallel shock absorbers must be considered. From Fig. 3, the below equation can be clearly obtained. Where F p represents the vector force applied by parallel shock absorber; F s represents the vector force applied by serial shock absorber; and F l represents the vector force applied by linear actuator. By measuring the pressure inside each air spring, the force applied on each shock absorber can be determined. Experiment and simulation demonstrate the characteristic of air spring as shown in Fig. 9. Despite the air spring has the generally unwanted property that the amount of compression and extension is not proportional to the force applied on it like the conventional spring, the force data can still be directly measured by pressure sensor. Fig. 9. The characteristic of air spring. As shown in Fig. 10, the dark blue line denoted series+parallel stands for the scale sum of the force applied by the both shock absorbers which completely overlap the red line, which represents the force applied by the linear actuator, this fact by the analysis of software in turn proof the above equation. Fig. 7 The simplified model of a single joint built in ADAMS Fig. 8 The diagram showing the linear relation of joint angle and actuator stroke. Fig. 10 The diagram showing that the vector sum of the force equals zero. 1168
C. Pictures Fig. 13 The captured pictures from the leg swing video. Fig. 11 The picture of frontal view of manufacture result. Fig. 14 Pitch joint angle of hip acquired by absolute rotary encoder. Fig. 12 The picture of rear view of manufacture result. D. Swing Test According to the periodic sequence aforementioned, a leg swing test is made as shown in Fig. 13. Below are some data acquired through testing process. The swing of pitch joint of hip is tested, Fig. 14 shows the data acquired from absolute rotary encoder. Fig. 15 shows the data from the internal incremental position sensor of the linear actuator. The rather linear relation between the actuator stroke and joint angle can be clearly seen. Fig. 16 shows the pressure data of the serial mounted shock absorbers of pitch joint of hip. It can be seen that the data is very unstable and sometimes below zero level. The probable reason is that the leg does not touch the ground, so the pressure change are tiny and can be viewed as chattering. Fig. 15 Hip actuator stroke acquire by the internal incremental position sensor. Fig. 16 Pressure data of serial shock absorber of hip acquired by pressure sensor. 1169
VI. CONCLUSIONS The implementation of the specifically designed mechanism for humanoid robot leg is presented in this paper. It features with the parallel and series shock absorbers and parallel linkage mechanism design. Instead of the conventional rotary motor used to propel humanoid biped walking robot designed with traditional method, the linear motion of shock absorbers and linear actuators mimics the movement of human muscle, thus bearing greater resemblance to the function of human muscle. The compliant device such as air spring in our case not only acts with the function of absorbing shock from external environment, but also provides the ability to assist the actuators to support the structure and utilizes the inertia of the humanoid biped walking robot legs itself to maintain the walking motion and reduce the output force requirement of the actuators. In other words, the lighter actuators of lower output force can be applied to sustain the whole humanoid biped walking robot that weighs much more than the maximum output force of actuators, which means the energy efficiency of the humanoid biped robot legs are also elevated to a level closer to human muscle while remain low cost since the mechanical requirement of the force transmission section is drastically decreased. The totally different paradigm of the whole new design means the algorithm adopted should also vary. The implementation of the control algorithm proposed in this paper is much simpler than traditional approach such as ZMP criterion. at the expense of the rather more complicated mechanism design than conventional humanoid biped walking robot. The control algorithm, Limit Cycle Walking, developed by Delft University of Technology, along with the Series Elastic Actuation introduced by Massachusetts Institute of Technology, is proven successfully achieve the result of natural human-alike walking by the active-actuated robot, Flame. The simulation result indicates that the robot leg can achieve the walking speed around 2.4km/h, about the same velocity as averaged children walk. The linear actuators utilized in our design are chosen under consideration in terms of power efficiency and cost, and thus the performance can merely meet the requirement. The walking speed can be significantly increased by exploiting the faster linear actuator such as linear motor. Furthermore, faster linear actuators not only improve the walking velocity, but also broaden the versatility since the capability of performing multi tasks such as jump or sprint is also a primary goal of our future research. Along with the programmable pneumatic pump, the spring constants and damping constants of shock absorbers can be online adjusted and in turn leads to the instantly varied support force contribution of all the shock absorbers mounted, the possibility of the specifically designed humanoid robot legs can reach is far beyond imagination. REFERENCES [1] K. Hirai et al. The development of Honda humanoid robot, in IEEE Proc. Robotics and Automation Conf., Leuven, 1998, pp.1321-1326 [2] Y. Sakagami et al. The intelligent ASIMO: system overview and integration, in IEEE Int. Conf Intelligent Robots and Systems, 2002, pp.2478-2483 [3] J. Chestnutt et al Footstep Planning for the Honda ASIMO Humanoid, in IEEE Proc Int Conf Robotics and Automation, 2005, pp. 629-634 [4] Hun-ok Lim and Atsuo Takanishi Biped walking robots created at Waseda University: WL and WABIAN family, in Phil. Trans. R. Soc. A 2007 365, 49-64 [5] Yu Ogura et al. Human-like Walking with Knee Stretched, Heelcontact and Toe-off Motion by a Humanoid Robot, in IEEE Int. conf. Intelligent Robots and Systems, Beijing, 2006, pp.3976-3981 [6] S. Lohmeier et al. System Design and Control of Anthropomorphic Walking Robot LOLA, in IEEE Trans Mechatronics, vol.14, no.6, pp.658-666, Dec. 2009 [7] S. Lohmeier et al. Modular joint design for performance enhanced humanoid robot LOLA, in IEEE Int. Conf. Robotics and Automation, Orlando, FL, 2006, pp.88-93 [8] S. Lohmeier et al. Leg Design for a Humanoid Walking Robot, in 6th IEEE-RAS Int. Conf. Humanoid Robots, Genova, 2006, pp.536-541 [9] D. Hobbelen et al. System overview of bipedal robots Flame and Tulip tailor made for limit cycle walking, in IEEE Int. Conf. Intelligent Robots and Systems, Nice, 2008, pp.2486-2491 [10] Jerry Pratt and Benjamin Krupp Series Elastic Actuators for legged robots, in Proc2004 SPIE. [11] S. Lohmeier et al."humanoid robot LOLA, in IEEE Int. Conf. Robotics and Automation, Kobe, 2009, pp.775-780 [12] Daan G.E. Hobbelen and Martijn Wisse Limit Cycle Walking, in Humanoid Robots: Human-like Machines, Vienna, Austria, I-Tech Education June, 2007, Ch. 14, pp. 277-294 [13] David Robinson et al. Series Elastic Actuator Development for a Biomimetic Walking Robot, in IEEE Int. Conf. Advanced Intelligent Mechatronics, Atlanta, GA, 1999, pp.561-568 [14] T. McGeer, Passive walking with knees in IEEE Proc Int. Conf. Robotics and Automation, Cincinnati, OH,1990, pp1640 [15] T. Buschmann et al. Biped walking control based on hybrid position/force control, in IEEE Int. Conf. Intelligent Robots and Systems, St. Louis, MO, 2009, pp.3019-3024 [16] Daan Hobbelen and Martijn Wisse A Disturbance Rejection Measure for Limit Cycle Walkers: The Gait Sensitivity Norm, in IEEE Trans. Robotics, vol. 23, no. 6, pp.1213-1224, Dec 2007 [17] Martijn Wisse et al. Adding an Upper Body to Passive Dynamic Walking Robots by Means of a Bisecting Hip Mechanism, in IEEE Trans. Robotics, vol.23, no. 1, pp.112-123, Feb. 2007 [18] Jerry Pratt, Ben Krupp, Design of a bipedal walking robot in Proc. 2008 SPIE [19] R.Luo, C.H.Chen, Y. Pu, J. Chang " Towards Active Actuated Natural Walking Humanoid Robot Legs,"in IEEE Int. Conf. on Advanced Intelligent Mechatronics (AIM 2011), Budapest, Hungary, 2011 1170