Bipedal Running Robot Using Pelvic Movement

WATHLETE-1

History
  1. Year 2013
  2. Year 2014
  3. Year 2015
  4. Year 2016
  5. Year 2017
  6. Year 2018

Year 2013

■ Running Model Using Pelvic Movement and Leg Elasticity

Fig. 1 SLIP2model
　The human running motion can be modeled by a spring-loaded inverted pendulum (SLIP). However, the SLIP model does not include the human-like pelvic movement which can contribute to increase a jumping force and absorb a landing impact. Based on a running gait analysis, we propose a novel model, named SLIP2 (Spring-Loaded Inverted Pendulum using Pelvis), that is composed of a body mass, a pelvis and a leg springs. This model can store elastic energy during running by utilizing pelvic movement in the frontal plane. Additionally, we developed pelvic oscillation control and running velocity control. To evaluate the proposed model and control methods, we performed hopping and running simulations. The simulation results show that the SLIP2 model achieved hopping and running motions. Moreover, it shows that the difference of the pelvic oscillation phase affects the jumping force．
○Simulation with SLIP2 model

[Related papers]
・Takuya Otani et al., "Utilization of Human-Like Pelvic Rotation for Running Robot," Front. Robot. AI Vol.2, No. 17, June, 2015. doi: 10.3389/frobt.2015.00017．

■ Pelvis
Pelvis CAD

Fig. 2 Mechanical design of the pelvis and hip joint
　From previous research, we know that a human running can be seen as a spring-mass model. Therefore, our Running Robot should use spring in its legs to use running dynamics and recycle kinematic energy.
　By analyzing human running, we find that the human pelvis movenent is similar to a sine wave in the frontal plane. Thus we develop a pelvis with DC servo motors and make it move sinusoidally. The leg with spring is then pushed off the ground the floor and the robot jumps.

■ Straight-Leg
The straight-legs and the pelvis

Fig. 3 The straight-legs
and the pelvis

Fig. 4 The sectional drawing of
the straight-leg

 　Previous research shows that a human running can be modeled as a mass-spring model whose stiffness changes as the running speed changes. To achieve this behaviour, we developed the Straight-Leg, which uses a spring (Fig. 3). 　The spring in the Straight-Leg (Fig. 4) can be changed in order to carry out experiments with different stiffness.

■ Hopping Motion Using Pelvic Movement and Leg Elasticity


　Fig. 5 Robot with a pelvis
　and a leg
  We describe a new hopping robot which can mimic the characteristics of human running. Based on a running gait analysis, we found that humans efficiently move the pelvis to utilize leg elasticity and increase jumping force. The hopping robot consists of a 1-DOF roll waist, a 1-DOF roll hip and a compression spring. Jumping experiments were conducted to verify the mechanisms of the robot. The robot achieved a 11 mm jump by using the pelvic rotation. We showed the effectiveness of the leg elasticity with pelvic rotation.
○Experiment
Mass ：　55[kg]
Leg stiffness ：　16[kN/m]
Oscillation period：　0.37[s]

[参考資料]
・Takuya OTANI et al., "Running Model and Hopping Robot Using Pelvic Movement and Leg Elasticity," Proceedings of the 2014 IEEE International Conference on Robotics and Automation (ICRA 2014), pp2313-2318, Hong Kong, China, May 2014.

■ Rotational Joint Leg with the Stiffness Module Using Leaf Spring
回転関節脚CAD図

Fig. 6 Rotational Joint Leg
弾性調節のしくみ

Fig. 7 Drawing of the stiffness module
　According to previous research, knee and ankle joints can be modeled as a torsion spring when human beings run and the stiffness changes as the running speed changes. Therefore, we have developed the Rotational Joint Leg with the stiffness module using leaf springs. 　The stiffness module (Fig. 7) uses leaf springs. The stiffness of the module depends on the adjustable roller, in other words, the length of the leaf spring. One stiffness module uses two leaf springs to cover wide range of stiffness.

■ Hopping Motion with Rotational Joint Leg

Fig. 8 Rotational joint legs
  Human running can be modeled as a mass-spring model, where the knee and ankle joints can be modeled as torsion springs. Thus we have developed the rotational joint legs using leaf spring. 　The leg has a stiffness module using leaf spring so that it simulates the stiffness of human running. Moreover, the robot with the leg can jump about 10 [mm] height when it is pushed from top cyclically.
○Experiment

[Related papers]
・Takuya OTANI et al.，"Leg with Rotational Joint That Mimics Elastic Characteristics of Human Leg in Running Stance Phase," Proceedings of the 14th IEEE-RAS International Conference on Humanoid Robots (Humanoids 2014),pp. 481-486，Madrid, Spain, November, 2014．

■ Verification of the Contribution of Pelvic Axial Movement to Running


Fig. 9 Pelvic axial rotation and hip torque
  We verify the contribution of pelvic axial movement to running using SLIP2 model. Pelvic axial rotation may help to increase the hip torque of the standing leg and running velocity. 　We develop the control method using pelvic axial rotation in order to simulate running.
○Simuration using pelvic axial rotation

[Related papers]
・ 大谷拓也他，“骨盤運動に着目した２足走行ロボットの開発　(第4報：　骨盤回旋運動を利用した走行制御)，”日本ロボット学会第31回記念学術講演会予稿集，1C1-03，Tokyo, Japan, 2013.
▲Go to the top of this page

Year 2014

■ Pelvis
腰部機構

Fig. 10 Pelvic mechanism

The pelvic mechanism developed in 2012 cannot do jumping motions which mimics the human's pelvic sine movement of 5 [deg] of amplitude at the target speed (4 [m/s]). Therefore, in the year 2013, we developed the new waist which improves the strength of the joints by reviewing the safety factor of the reduction gears (Harmonic Drive). Furthermore, the pelvic mechanism has the same distance between both hip joints as human’s.



■ Continuous hopping motion with landing estimation
　The simulation shows that the motion of the center of gravity (CoG) and the pelvic phase shift that occurs during the flight phase reduce the height of the jumping. Hence we equipped the robot with an oscillation control that reduces phase shift by estimating stance phase. Consequently, the Running Robot can jump continuously without decreasing the height of the jumping.
　○Continuous jumping experiment

[Related papers]
・ Takuya Otani et al., "Utilization of Human-Like Pelvic Rotation for Running Robot," Front. Robot. AI Vol.2, No. 17, June, 2015. doi: 10.3389/frobt.2015.00017．

■ Development of Running Robot with knee and ankle joints
　According to previous research, human running has the following characteristics.
   1. Knee and ankle joints can be modeled as torsion springs in stance phase.
   2. Joint stiffness changes over running speed.
   3. Knee joint moves at 6.5[rad/s] in swing phase.
   4. Knee joint withstands torque of 177[Nm] in stance phase.
　The Running Robot has leaf springs on the knee and ankle joints in order to mimic characteristic 1. (Fig. 11)．Also, the Robot has a Joint Stiffness Adjustment Mechanism on the knee joints to mimic thecharacteristic 2. (Fig. 12)．
　It is difficult to develop a compact knee joint mechanism using a DC motor and a reduction gear for mimicking characteristics 3. and 4. So, we used the self-locking function of worm gears to develop compact Knee Joint Mechanisms (Fig. 13)．

Fig. 11 Running Robot with knee and ankle joints

Fig. 12 Joint Stiffness Adjustment Mechanism

Fig. 13 Knee Joint Mechanism

[Related papers]
・ Takuya Otani, et al., "Knee Joint Mechanism That Mimics Elastic Characteristics and Bending in Human Running," Proceedings of the 2015 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2015),pp. 3969-3974，Hamburg, Germany, September, 2015．
▲Go to the top of this page

Year 2015

■ Running velocity control by changing foot placement
　When robot change joint angles in case of the multi-joint leg, ground reaction force vector also change. If the vector doesn't point centor of mass, it is difficult to jump highly and to run stably. So, we estimated ground reaction force vector with multi-joints leg model. As a result, we can estimate joint angles when ground reaction force points centor of mass. Moreover, we enabled robot to control running velocity by changing foot placement.
○Hopping experiment

[Related papers]
・ Takuya OTANI et al., "Running with Lower-Body Robot that Mimics Joint Stiffness of Humans," Proceedings of the 2015 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2015), pp. 3969-3974, Hamburg, Germany, September 2015.

■ Development of the new active knee mechanism
　We succeeded in hopping with one leg by pelvic movement and leg elasticity. Then, we developed active knee mechanism to push off the ground actively in order to run faster and to improve power (Fig. 14). We used backdrivable worm gear to develop higher power knee mechanism. It is higher to transmit torque from DC motor, but it is lower to transmit torque from ground.
　Moreover, we used CFRP leaf spring (density: 1.5[g/cm2]) for mimicking joint stiffness because steel leaf spring is very heavy (density: 7.8[g/cm2]). However, one leaf spring is strong poverty because of stress concentration. So, we used laminated leaf spring in order to decrease stress concentration. As a result, we developed high power knee joint mechanism which is similar size and weight to human.
　We succeeded in starting jumping by one active push off.
○Jumping experiment

Fig. 14 knee mechanism

[Related papers]
・Takuya Otani, et al., "Joint Mechanism That Mimics Elastic Characteristics in Human Running," Machines 2016, 4(1), 5, 2016.

Year 2016

■ The angular momentum control
　Humans compensate the angular momentum of lower-limbs (Yaw dirention) by moving the upper body while running. Mimicking this feature is important to realize steady running. So, we should get the value of the angular momentum of lower-limbs in order to decide the monement of the upper body. Therefore, we developed the angular momentum control that is able to decide the movement of the upper body. This control calculates the angular momentum of lower-limbs and decides the angular momentum of upper body that is able to compensate the angular momentum of lower-limbs. By using this control, we comfirmed that the upper body is able to compensate the angular momentum of lower-limbs (Yaw direction).
○Compensating the angular momentum by the upper body

[Related papers]
・Takuya Otani, et al., "Angular Momentum Compensation in Yaw Direction using Upper Body based on Human Running," Proceedings of the 2017 IEEE International Conference on Robotics and Automation (ICRA 2017), pp. 4768-4775, 2017.

■ Development of the upper body mechanism having human's mass properties and capable of generating human's angular momentum
　We have already succeeded in hopping with one leg in a state restricted by a guide. However, the runnnig robot didn't hane the upper body, so the robot was not able to mimic the above features of human's upper body. In particular, the feature that humans compensate the augular momentum of the lower limbs by moving upper boby is important in order to run steadily like humans. So, we developed the upper body mechanism capable of doing running movement(5.0[m/s]) and having human's mass properties and link lengths to generate the human's angular momentum.
　Upper body mechanism is showm in Fig. 7, Fig. 8 and Fig. 9. We adopted the sirial-link mechanism using Brushless motor(light, high power) to arm mechanism in order to realize the fast angular velocity and large range of motion. In addition, We adopted CFRP(Carbon Fiber Reinforced Plastics) to the robot frame in order to the requirement specification of mass. As for the trunk mechanism, We adopted the CFRP pipe to the robot frame in order to make robot lighter. Moreover, as you can see in Figure 10, the CoM of trunk must be in the center of the lateral direction in order to prevent the robot tilting during jumping. So, We developed the double motor mechanism moving trunk pitch axis using the same two motors. Finally, We developed the upper body mechanism having human's mass properties and link lengths.
　Moreover, We comfirmed that the upper body was able to generate human's angular momentum and able to compensate the angular momentum of lower-limbs.

Fig. 15 Upper body mechanism

Fig. 16 Arm mechanism

Fig. 17 Trunk mechanism

Fig. 18 Double motor mechanism（Trunk pitch-axis）

○Mesuring the angular momentum of the upper body

[Related papers]
・ Takuya Otani, et al., "Upper-Body Control and Mechanism of Humanoids to Compensate for Angular Momentum in the Yaw Direction Based on Human Running," Appl. Sci., 8, 44 2018.

■ Development of the new active ankle-foot mechanism capable of ground contact adaptation
　According to previous research, human running has the following characteristics.
   1. Knee and ankle joints can be modeled as torsion springs in stance phase.
   2. Joint stiffness changes over running speed.
   3. Ankle joint moves at 4.6[rad/s] in swing phase.
   4. Ankle joint withstands torque of 180[Nm] in stance phase.
   5. Ankle joint and foot move to adapt with the ground in contact phase.

　We have already succeeded in hopping with one leg by pelvic movement and leg elasticity. However the robot couldn't drive its ankle joints like a human and its standing posture was unstable due to foot-ground contact condition. Then, we developed a new ankle-foot mechanism to behave humanlike ankle joint motion during running; we used back drivable worm gear, which is also used on Knee joints, and two BLDC motors to develop higher power ankle mechanism and to withstand the shock from the ground.
　Mounting two BLDC motors around ankle joint makes the size of mechanism to be larger and the lower leg mass to be heavier. To solve this problem, we used the four-bar mechanism to transmit torque from one motor, which is mounted below the knee, to ankle joint and the other motor is mounted at near the ankle; thus we developed high power ankle joint mechanism which is similar size and weight to human. This mechanism also has three point ground contact parts on the sole of robot foot, which allow the robot to stand stably.
　Moreover, we discovered the new feature of running movement due to the analyzing of human running. During running, especially at the ground contact phase, humans land the outside of foot first, and move ankle joint and foot to adapt with the ground. For mimicking this feature, we used CFRP leaf spring and develop the mechanism not only to bend but also to twist them; thus we succeeded in mimicking joint stiffness and ground adaptation motion during jumping.

Fig. 19 Ankle-foot mechanism

Fig. 20 Ankle-foot mechanism (pitch-axis)

Fig. 21 Ankle-foot mechanism (roll-axis)"

○Jumping Experiment with Ground Adaptation Motion

Fig. 22 Ground adaptation motion

[Related papers]
・ 夏原他，"骨盤運動に着目した２足走行ロボットの開発 (第14報： 弾性要素を有し能動動作と路面への倣い動作が可能な足関節・足部機構)，"日本ロボット学会第34回記念学術講演会予稿集，RSJ2016AC3Y2-01，山形県，2016年9月．
▲TOPへ

Year 2017

■ Trunk joint control considering angular momentum
　Various stabilizing control methods for humanoids during the stance phase while hopping and running have been proposed. Although these methods contribute to stability while hopping and running, it is possible that the control during the flight phase could also affect the stability. In this study, we investigated whether the control during the flight phase can affect the stability of a humanoid while running. To achieve stable hopping, we developed a control system that accounts for the angular momentum of the whole body during the flight phase. In this system, the angular momentum generated by the motion of the lower body in each time interval is calculated during the flight phase, and the trunk joints are controlled to generate the angular momentum necessary to compensate for the deviation of the waist posture, which is used as the reference point for the motion coordinate system of the robot. Once the proposed control system was developed and simulated, we found that the hopping duration in the unconstrained state was extended.
○Simulation with trunk joint control

[Related papers]
・ Takuya Otani, et al., "Trunk motion control during the flight phase while hopping considering angular momentum of a humanoid," Advanced Robotics, 32, 8, 2018.

■ Variable joint stiffness mechanism for transition between low-speed and high-speed running
In previous studies, human running has the following characteristics.
   1. Knee and ankle joints can be modeled as torsion springs in stance phase.
   2. Joint stiffness changes over running speed
   3. Joints withstand large torques as high as 180[Nm] in stance phase.


In our previous researches, we have developed the joint mechanism that mimics human's joint stiffness and can output torques as high as those generated during human running. However, the mechanism does not mimic the joint stiffness needed for low-speed running of human. In this study, we developed a new joint mechanism that has a wide joint stifness range which corresponds to the stifness range between low and high speed running, and can change its joint stifness between the minimum value and the maximum value during the flight phase. At first, we determined the requirements of the joint mechanism based on the previous studies in sports sciences. About the joint stiffness for running, the minimum requirement is 320[Nm/rad] which corresponds to running at a speed of 2.0[m/s], and the maximum requirement is 790[Nm/rad] which corresponds to running at 6.6[m/s]. Moreover, if the joint could increase its stiffness as high as that needed for conventional walking, the mechanism could be used not only for running, but also for walking. Therefore, we determined the maximum requirement of the joint stiffness as 3000[Nm/rad] on the basis of the hip roll joint stiffness of WABIAN-2R.
For achieving these requirements, a trapozoidal laminated CFRP leaf spring is implemented in place of a rectangular laminated CFRP leaf spring for decreasing the minimum joint stiffness (Fig. 23). To change the effective length of the leaf spring to be able to attain the maximum joint stiffness, a system to change the position of the fixed point instead of the loading point of the leaf spring was implemented. To move the fixed point, a ball screw and a heavy duty linear guide are implemented in the joint (Fig. 24). From the evaluation experiment with the developed mechanism, it is confirmed that the developed mechanism fulfills the requirements about the range of joint stiffness, torque capacity, and changing speed of joint stiffness.
We plan to implement the mechanism into WATHLETE-1 and realize a transition between low-speed and high-speed running.

Fig. 23 Rectangular leaf spring and trapezoidal leaf spring


Fig. 24 Variable joint stiffness mechanism with the system to move the fixed point of the leaf spring

[Related papers]
・ 赤堀孝太他，"骨盤運動に着目した２足走行ロボットの開発（第16報：広範囲剛性関節機構のための台形 CFRP 重ね板ばね），"日本ロボット学会第35回記念学術講演会予稿集，3L1-04，埼玉県，2017年9月．
・ 尾原睦月他，"骨盤運動に着目した２足走行ロボットの開発（第17報：低速走行から高速走行への遷移に対応した広範囲剛性可変機構），"日本ロボット学会第35回記念学術講演会予稿集，2L1-03，埼玉県，2017年9月．
▲TOPへ

Year 2018

■ Achievement of jumping motion utilizing actuators and elastic components
In previous studies, we developed joint mechanism which has actuators and elastic components. But, when mechanism move by energy accumulated in elastic components, actuator must output same torque as elastic components output because actuators and elastic components are connected in series. So, in flight phase, we perform joint position control by actuators, in stance phase, joints move by elastic components only.
In steady running, robot can accumulate energy by converting potential energy to elastic energy. But when robot start jumping from standing position, robot must output energy from actuators and elastic components because elastic components can’t accumulate enough energy.
So, we achieve jumping motion with reference to countermovement jump (Fig. 7). To achieve this movement, we perform the following process (Fig. 8). First, we generate CoM (center of moment) and foot motion considering elastic components. Next, we generate joint motion from CoM and foot motion. Finally, we generate actuator motions of each joint by using dynamic equation.

Fig. 7 Outline drawing of countermovement jump

Fig. 8 Jumpping motion generator considering joint elasticity

[Related papers]
・ Takuya Otani, et al., "Jumping Motion Generation of a Humanoid Robot Utilizing Human-like Joint Elasticity," Proceedings of the 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2018),pp. 8707-8714，Madrid, Spain, October, 2018．


■ Variable joint stiffness mechanism for transition between low-speed and high-speed running
In previous studies, human running has the following characteristics.
   1. Knee and ankle joints can be modeled as torsion springs in stance phase.
   2. Joint stiffness changes over running speed
   3. Joints withstand large torques as high as 190[Nm] in stance phase.
   4. Outputting energy is larger than accumulated energy when standing phase in running at ankle joint.

In previous studies, we achieved same quasi joint quasi stiffness and output as humans at knee joint. However, in ankle joint, we only achieved same quasi joint quasi stiffness as human running fast, and the robot’s ankle joint can’t move actively when running standing phase. So, we developed A new ankle mechanism capable of kicking at jumping and running and adaptable to change of running speed. At first, we determined the requirements of the joint mechanism based on the previous studies in sports sciences. The joint needs to kick at 190[Nm], the max torque at running standing phase, to kick actively. And about the joint stiffness for running, the minimum requirement is 250[Nm/rad] which corresponds to running at a speed of 2.0[m/s], and the maximum requirement is 325[Nm/rad] which corresponds to running at 5.0[m/s].
For achieving these requirements, the ankle pitch joint’s actuator, which kick with high power when running, change from single motor to double motor (Fig. 9). And about the elastic components, trapezoidal and rectangle laminated CFRP leaf springs are implemented in series at ankle joint to achieve low joint quasi stiffness (Fig. 10). To implement Variable joint quasi stiffness mechanism, the pitch joint’s actuator is mounted at upper lower leg and transmit power with 4-bar linkage. At foot, to make sure the axis does not shift, the leaf spring is connected to ankle joint with 4-bar linkage, because if the axis of mechanism and the axis of leaf spring are different, joint quasi stiffness becomes non-linear (Fig. 11). From the evaluation experiment with the developed mechanism (Fig. 12), it is confirmed that the developed mechanism fulfills the requirements about the range of joint stiffness, but it doesn’t satisfy output torque. The reason why torque is lower than requirements is the high power motor driver is not able to use at experiment.
 　

Fig. 9 double motor


Fig. 10 CFRP trapezoidal leaf spring and Rectangular leaf spring

Fig. 11 Ankle mechanism links

Fig. 12 Ankle mechanism

[Related papers]
・Hiroki Mineshita, Takuya Otani, Kenji Hashimoto, Masanori Sakaguchi, Yasuo Kawakami, Hun-ok Lim and Atsuo Takanishi, “Robotic Ankle Mechanism Capable of Kicking While Jumping and Running and Adaptable to Change in Running Speed,” Proceedings of the IEEE-RAS International Conference on Humanoid Robots (Humanoids 2019), pp. 529-534, Tront, Canada, October, 2019.
▲TOPへ

Last Update: 2019-Nov
Copyright(C) 2019 Takanishi Laboratory
All Rights Reserved.