Skip to main content

Realization of Natural Human Motion on a 3D Biped Robot For Studying the Exoskeleton Effective



In this study, natural human motion was modeled using a three-dimensional simulation involving a biped robot. Exoskeleton assistance was examined through the extraction and analysis of kinematic and dynamic parameters. The present findings can serve as a reference for a study on exoskeleton design in which user effort is considered.


A biped robot simulator of human gait was constructed. A participant’s movement was recorded using a Vicon motion capture system. The effect of exoskeleton assistance on gait performance was evaluated under admittance control for user interaction.


In the simulation of a squatting motion, the exoskeleton helped the user lift a 50-lb weight without the user exerting any additional effort. Exoskeleton energy consumption was also examined. Virtual parallel bars effectively assisted the biped in simulating natural walking gait. Adjusting the body’s center of gravity helped reduce the robot’s dependence on the parallel bars, as did changing the walking speed, which allowed the body to catch up with leg motion.


The biped effectively simulated the participant’s motion when he did not walk away from where he stood. Special care was taken because the biped robot lacked the degrees of freedom of human motion. Notably, virtual parallel bars were successfully used to help the robot walk with a natural gait. This method can be applied in scenarios mimicking humans moving with assistance.


In this study, natural human motion was simulated using a three-dimensional biped robot to aid in the development of exoskeleton design through the extraction and analysis of joint parameters. The study of human pilot joint forces depends almost entirely on musculoskeletal modeling and dynamic simulation. Such simulation models typically do not consider the impacts of the exoskeleton. In other words, the joint force of the human user under the influence of the skeleton cannot be derived. Biped robots, which realistically simulate lower-limb movement, are attracting increasing scholarly attention.

The present study investigated the effort exerted by exoskeleton users in motion. Human exoskeletons were invented in the 1960s [1]. Today, numerous commercial active exoskeleton systems are available, including the Hybrid Assistive Limb by Cyberdyne Inc. [2], ReWalk by ReWalk Robotics [3], EKSO by EKSO Bionics [4], and the Indego Therapy Exoskeleton [5], all of which offer assistance with walking. Passive exoskeletons, which are designed to add to the user’s power during execution of a movement, include the Human Augmentation Research and Development Investigation Manipulation exoskeleton [6] and the Berkeley Lower Extremity Exoskeleton [7]. Regarding the aspects of control and design, Strausser and Kazerooni proposed using a humanoid machine interface to help the user drive the sit-to-stand motion. This review article examined the trends and challenges in exoskeleton design [8]. Studies have also assessed lightweight, passive assistive devices that are not burdened by actuators and fuel packs [9,10,11]. Notably, the Massachusetts Institute of Technology Media Laboratory developed a system on the basis of the concept of passive assistive devices [12]. Regarding motion control, Dariush [13] proposed an approach for separating the mechanism involved in voluntary motor control and assistive exoskeleton control. Other researchers [14] studied a passive exoskeleton with artificial tendons, noting that it provided less assistance than expected. Another study [15] presented a weight-supporting biped exoskeleton that constituted a quasi-passive assistive device. Regarding modeling, most studies have used two-dimensional (2D) models to describe exoskeleton dynamics [16,17,18,19]. In one study [19], a realistic 2D model was used to determine the effects of a quasi-passive, energy-efficient power-assisted device. The present study focused on modeling user movements to examine exoskeleton assistance performance. The model must be able to both mimic human walking and take into account the efforts exerted by the exoskeleton. Although many studies have been conducted on human exoskeletons, very few investigations have considered human gait. Aoustin and Formalskii [20] modeled human walking by using a planar bipedal anthropomorphic mechanism. Although they used a 2D model, they were able to implement walking controls on a three-dimensional (3D) biped robot. Lim et al. [21] simulated human movement on a 3D biped robot using motion data captured from human participants, imposing numerous constraints on the hip center, the supporting foot, and the swing foot in the 2D simulation. The study did not consider exoskeletons. Chen et al. [22] proposed a gait planning approach for assisting walking through the provision of balance and gait stability. Simulations were performed on a 3D biped robot, with crutches used to maintain balance.

The simulation of natural human gait on a biped robot wearing an exoskeleton allows the examination of the device’s assistive effect. Developing a biped model that reflects the human skeleton and human motion is incredibly challenging. Although squatting motions can be adequately modeled to determine exoskeleton assistance, simulating human gait in 3D biped robots is almost impossible because of human–robot mismatches. To resolve this problem, we used a virtual parallel bar to assist the robot in modeling human gait, which allowed us to evaluate exoskeleton assistance.

In the present study, a motion capture system sent data into an OpenSim simulation for calculation of the hip, knee, and ankle angles. The joint angle variations or the human motion profiles were then used to drive a biped robot, which acted autonomously in the gait simulation. A virtual exoskeleton was attached to the robot, and its assistive effect was determined. The joint forces and acceleration of various parts of the body were then extracted. This information can be used for the identification of human motion intention in future studies.

Human Motion Capture

A Vicon motion capture system was used to record a human participant’s movement patterns (Fig. 1a). We then developed a dynamic simulation model of human motion in OpenSim (Fig. 1b). The blue and red dots indicate the locations of motion capture beads. This model mapped motion data onto limb postures. The limb movement profiles enabled the calculation of the joint angles by OpenSim, which were in turn used in the biped simulation.

Fig. 1

a Human motion capture experiment and b three-dimensional dynamic simulation model of human motion built using OpenSim software

Squatting Motion

The squatting and walking motions of the participant were captured and analyzed. Specifically, the squat-to-stand movements of the participant carrying loads of 0, 25, and 50 lbs were captured. Figure 2 shows the angular movements of the knee, hip, and ankle during the standing part of the squat-to-stand movement. Initially large in value, the angles gradually declined to zero as the participant reverted to a standing pose.

Fig. 2

Participant’s joint angles during squatting, presented in OpenSim

Walking Motion

Figure 3a displays the patterns in the angular changes of walking gait.

Fig. 3

a Joint angle variations of human walking gait. b Simulated joint angles

Because the orientation of the biped coordinates was opposite to that of the motion capture system, the signals had to be reversed; two large angular peaks were observable in the knee bend (Fig. 3b). The first small bend occurred in response to the shifting of body weight to the stance leg. To compensate for human–robot differences, we eliminated this minor bend in the simulation because it caused the robot to fall. This smaller hump indicates that the stance leg’s knee joint will slightly bend and squat during the natural human gait. In the simulation, we only considered the change in the sagittal plane, hence no lateral movement. If the stance leg bends downward in these conditions, the swing leg will kick the ground to fall. Figure 4 shows the corresponding gait changes.

Fig. 4

Modified walking gait patterns in the robot simulation

For the simulation, we partitioned the gait into the stance phase and the swing phase. During the stance phase, the stance leg moves with the knee extended straight, as indicated by the change from posture A to posture B in Fig. 5. During the swing phase, the heel lifts off while the knee bends toward the ground, generating a large angular motion. The knee then straightens, allowing the heel to touch the ground (Fig. 6).

Fig. 5

Stance phase of the gait cycle

Fig. 6

Swing phase of the gait cycle

During the stance phase, the stance leg’s ankle is viewed as a pivot and the center of gravity of the body is above the waist. Although the ankle produces small movements, it can exert substantial postural changes because of its considerable distance from the waist. The joint angle trajectories provided the basis for the robot simulation.

Synchronous Control of Biped Motion Under Exoskeleton Assistance

The effect of knee movement on human movement under exoskeleton assistance is discussed as follows.

Single-Axis Exoskeleton Control

We first established a stable controller for a single exoskeleton and then extended the results to achieve synchronized motion in both knees, upon which human movement could be incorporated in studying the assistive effect of the exoskeleton. The single-joint exoskeleton control system, which was based on an impedance control algorithm, consisted of a robust inner loop for motor control and an outer loop for admittance control. In favor of brevity, the motor control is not discussed. As shown in Fig. 7, \({T}_{d}\) and \({T}_{m}\) be the user command and control torque acting on Pilot Simulation. The joint velocity \({\omega }_{m}\) and joint torque \({T}_{total}\) will be measured. The difference between \({T}_{total}\) and \({T}_{m}\) is the estimated pilot command \({\widehat{T}}_{d}\), which is then input the admittance controller to compute the desired velocity \({\omega }_{d}\). In response to the motor speed \({\omega }_{c}\), the exoskeleton moter reflects the control torque [23]. \({K}_{a}\) is the gain of the admittance controller, and \(a\) is the corner frequency.

Fig. 7

Single-joint exoskeleton control algorithm

As shown in Fig. 6, the swing leg is supported by the stance leg during walking. Parameter \({K}_{a}\) controls the amount of effort exerted by the exoskeleton. Figure 8a and b show the simulated torque responses of the admittance controller under the effects of \(a\) and \({K}_{a}\), respectively. The lower corner frequency and the higher gain both resulted in more prominent actuator responses. Notably, user torque was enhanced almost threefold.

Fig. 8

a Effect of \(a\) on the admittance controller. b Effect of \({K}_{a}\) on the admittance controller

Synchronized control of both knees was required to ensure that the robot walked in a natural manner. Squatting and walking motions were examined in the synchronized control to determine the assistive effect of the exoskeleton on robot motion.

Robotic Simulation of Natural Human Movement

The limb movement profiles enabled the calculation of the joint angles by OpenSim, which were in turn used in the biped simulation. The biped robot, which was modeled by Matlab Simulink Simscapes, was designed to match the characteristics of the participant (Fig. 9), had a height and weight of 181 cm and 97 kg, respectively. The biped robot is a six-degrees-of-freedom model included a torso and two legs. The each joints (hip, knee, and ankle) can only act in the sagittal plane and be constrained in specific angle interval to mimic the human motion.

Fig. 9

Biped robot modeling


The underlining exoskeleton was designed to assist in tasks involving heavy lifting. In the first part of the simulation, the robot lifted objects weighing 0, 25, and 50 lbs. Exoskeleton assistance was examined after the exoskeleton was incorporated into the simulation. The motion capture system only captured data from the standing phase. In this test, we used half-sinusoidal signals to approximate the joint trajectory to present a complete squatting cycle (Fig. 10).

Fig. 10

Robot performing a squatting motion

Squatting motion tests were first conducted with the biped carrying 0-, 25-, and 50-lb loads (Fig. 11a, b). Heavier loads led to the proportional production of hip and knee torque. As shown in Fig. 11a and b, the movement comprised two components: a shift in torque bias caused by the weight and a torque component to drive the corresponding motion of the moving load.

Fig. 11

a Hip torque and b knee torque during squatting

We input the exoskeleton torque into each corresponding joint to examine exoskeleton assistance. Figure 7 shows the exoskeleton under admittance control. The driving torque of the user under no load, \({T}_{d0}\), was incorporated into the control loop. The admittance controller then drove the exoskeleton to comply with the driving torque requirements. The controller pole location \(a\) helped eliminate the noise signal induced by body sway. Next, the magnification of the torque was determined according to the controller gain \({K}_{a}\). The controller output was then sent to the motor driver of the exoskeleton to assist in the squatting motion.

As shown in Fig. 12, under the same \({T}_{d0}\), the biped successfully lifted the 25- and 50-lb loads, with the exoskeleton providing the additional torque. Similar observations were made for the hips and knees.

Fig. 12

a Hip and b knee torque under exoskeleton assistance

Estimation of Energy Consumption for Squatting

Still in the simulation stage and having not yet obtained the experimental suit, we incorporated the weight of the suit (including the battery pack, motor, and controller) into the exoskeleton weight.

The following formula was used to calculate the energy expenditure:

$${W}_{m}={\int }_{0}^{t}\left|\left[{T}_{total}\left(t\right)-{T}_{d}\left(t\right)\right]{\dot{\theta }}_{human}(t)\right|dt$$

where \({W}_{m}\) is the motor power consumption, \({T}_{total}(t)\) is the torque generated by the motor, \({T}_{d}\) is the applied driving torque of the user, and \({\dot{\theta }}_{human}\) is the joint angular velocity. Figure 13a and b show the variations in the power consumption of the hip and knee joints (motors) of the exoskeleton, respectively. The two peaks in the power output correspond to the standing and squatting phases of the motion.

Fig. 13

Power consumption by the a hip and b knee joints of the exoskeleton

Let \({W}_{m}\) represent the total power consumed by the hip and knee motors. Figure 14 shows the exoskeleton’s total power consumption in the squatting motion, and Table 1 shows the power exerted by the user and the exoskeleton motor. The user produced 440 J, with the exoskeleton providing 271 and 543 J in the lifting of the 25- and 50-lb loads, respectively. As the load increased, the exoskeleton exerted an increased amount of force to help lift it.

Fig. 14

Total power consumption of the exoskeleton

Table 1 Total power consumption of the exoskeleton

Biped Walking

Floor resistance was first removed to allow the biped to walk in place, whereupon it began rocking and failed to maintain a straight path. As mentioned, building a 3D biped robot that accurately reflects the structure of the human skeleton is extremely difficult; human–robot mismatches preclude the biped from replicating human gait. Previous studies have only achieved or simulated such replications in 2D, specifically by restricting robot motion. After numerous trials, we concluded that only bipeds with very low mass would be able maintain a natural walking gait in 3D simulations. To resolve this problem, we used virtual parallel bars to help the robot walk along a straight line while retaining the 3D characteristics of the simulation.

Virtual Parallel Bars

The biped had the same weight and same center of mass as the participant. As shown in Fig. 15, we used virtual parallel bars to assist the biped in walking; specifically, a ball was hung inside the body on the desired straight line to mimic the effect of the bars. The ball only had a vertical supporting force (i.e., in the upward direction); to prevent the bars from affecting the robot’s forward movement, no resistance in the \(x\) or \(y\) directions was involved.

Fig. 15

Parallel bars used to support robot walking

The motion capture data revealed a slight bend in the stance leg during the stance phase that introduced instability into the biped simulation; therefore, we straightened the stance leg to support the swing leg. The same procedure was applied in [24, 25]. When people walk, they typically lean forward before stepping out in response to inertia. By contrast, the push tended to cause the biped to accelerate its gait, causing the body to lag behind and exhibit a slightly tilted gait (Fig. 16).

Fig. 16

Robot walking gait

The joint torque was extracted from the simulation after the input of the hip, knee, and ankle trajectories used to drive the biped. The extracted signals were contaminated by the body twist and the impact of the heel and toe on the floor; nevertheless, it was possible to discern a periodic pattern (Fig. 17). To reduce the noise, the frequency of the low-pass filter in the admittance controller was tuned, and the torque signals and the position signals were then input into the admittance controller to generate the assistive force of the exoskeleton (Fig. 18).

Fig. 17

Noisy joint torque feedback from the biped simulation

Fig. 18

Hip and knee torque under admittance control

Modified Parallel Bars

The parallel bars imposed a rigid constraint on the center of mass, resulting in an unnatural gait. To relax this constraint, we added vertical spring and damper forces to the center ball and allowed the robot to move around it. The robot could still move freely along the walking direction. The body velocity was initially set to 0.5 m/s. Again, the robot could not catch up as the lag increased, relying on the spring force to maintain balance. Figure 19 shows the resulting spring force and the forward speed variation. The average spring force after the motion had settled was 758 N, indicating that the parallel bars had borne most of the weight. The paddling generated forward thrust and caused a gradual acceleration in body velocity to an average of 1.7 m/s. The spring force oscillated vigorously before slowly settling into a small oscillation as the speed approached a steady-state value.

Fig. 19

a Spring force from the parallel bars. b Forward velocity variation

We conducted various tests to reduce the spring force. One attempt involved moving the center of gravity forward through the addition of an object of low mass (0.1 kg) in front of the body (Fig. 20). The effect of the center of gravity was then examined through gradual increases in the distance \(L\) from the body.

Fig. 20

Adjustment of the center of gravity to balance the body tilt

Table 2 summarizes the results.

Table 2 Effect of shifting the center of gravity forward

Table 2 shows that shifting the center of gravity forward reduced the backward lean. However, this effect was no longer observable when \(L\) reached 900 cm, at which point the amplitude of the forward/backward rocking motion had become too large to allow the maintenance of a steady gait, causing an increase in spring force dependence. Notably, when \(L=100\) cm, the violent rocking caused the biped to stumble.

From the velocity responses, it can be inferred that adjusting the walking speed mitigated the backward leaning problem. Fixing \( L\,{\text{at}}\,500 \) cm, we performed walking experiments with varying gait periods and observed the corresponding variation in the spring force.

Table 3 Effect of walking speed

Table 3 shows that smaller gait periods, or rapid walking pace, also reduced the spring force dependence. However, when the walking speed was too fast to allow the body to catch up, the robot fell again.


In the present study, natural human motion was simulated using a biped robot. Because all the biped motion parameters were available, data on velocity, acceleration, and even jerk were extracted from the simulation and used to analyze movement intention. The robot was used to examine the assistive effect of an exoskeleton. Specifically, we built a robot—fitted with an exoskeleton under admittance control—that can model human movement through 3D simulation. In the squatting experiment, the admittance control helped a participant to successfully lift a 50-lb load without exerting additional effort. The simulation also enabled the determination of exoskeleton energy consumption. Modeling natural human gait required additional treatment; a virtual-parallel-bars mechanism was implemented to assist the robot in walking. Adjusting the body’s center of gravity allowed the robot to reduce its dependence on the parallel bars. In addition, it helped adjust the walking speed to allow the body to catch up.

Data availability

The human gait data can be downloaded from the following server:

Code availability

The system is developed in MATLAB.


  1. 1.

    Dollar, A. M., & Herr, H. (2008). Lower extremity exoskeletons and active orthoses: Challenges and state-of-the-art. IEEE Transactions on Robotics, 24(1), 144–158.

    Article  Google Scholar 

  2. 2.

    Kawamoto, H., Hayashi T., Sakurai, T., Eguchi, K., Sankai, Y. (2009) Development of single leg version of HAL for hemiplegia. In: Proceedings of the 31st Annual International Conference of the IEEE Engineering in Medicine and Biology Society: Engineering the Future of Biomedicine, EMBC 2009, pp 5038–5043.

  3. 3.

    Talaty, M., & Esquenazi, A. (2013). Briceno JE differentiating ability in users of the ReWalkTM powered exoskeleton: An analysis of walking kinematics. IEEE International Conference on Rehabilitation Robotics.

    Article  PubMed  Google Scholar 

  4. 4.

    Swank, C., Sikka, S., Driver, S., Bennett, M., & Callender, L. (2020). Feasibility of integrating robotic exoskeleton gait training in inpatient rehabilitation. Disability and Rehabilitation: Assistive Technology, 15(4), 409–417.

    Article  PubMed  Google Scholar 

  5. 5.

    Farris, R. J., Quintero, H. A., Murray, S. A., Ha, K. H., Hartigan, C., & Goldfarb, M. (2014). A preliminary assessment of legged mobility provided by a lower limb exoskeleton for persons with paraplegia. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 22(3), 482–490.

    Article  PubMed  Google Scholar 

  6. 6.

    Rupala BS, Singla AS, Virk GS (2016) Lower limb exoskeletons: a brief review. Paper presented at the IJSRD Conference on Mechanical Engieering & Technology (COMET), Varanasi, Pradesh, Utter, 1/2016

  7. 7.

    Kazerooni, H. (2007). Human augmentation and exoskeleton systems in Berkeley. International Journal of Humanoid Robotics, 4(3), 575–605.

    Article  Google Scholar 

  8. 8.

    Yan, T., Cempini, M., Oddo, C. M., & Vitiello, N. (2015). Review of assistive strategies in powered lower-limb orthoses and exoskeletons. Robotics and Autonomous Systems, 64, 120–136.

    Article  Google Scholar 

  9. 9.

    van den Bogert, A. J. (2003). Exotendons for assistance of human locomotion. BioMedical Engineering Online.

    Article  PubMed  PubMed Central  Google Scholar 

  10. 10.

    Agrawal SK, Banala SK, Fattah A, Scholz JP, Krishnamoorthy V, Hsu WL A gravity balancing passive exoskeleton for the human leg. In: Robotics: Science and Systems, 2007. pp 185–189

  11. 11.

    Agrawal, S. K., Banala, S. K., Fattah, A., Sangwan, V., Krishnamoorthy, V., Scholz, J. P., & Hsu, W. L. (2007). Assessment of motion of a swing leg and gait rehabilitation with a gravity balancing exoskeleton. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 15(3), 410–420.

    Article  PubMed  Google Scholar 

  12. 12.

    Walsh, C. J., Pasch, K., Herr, H., (2006) An autonomous, underactuated exoskeleton for load-carrying augmentation. In: 2006 IEEE/RSJ International Conference on Intelligent Robots and Systems, pp 1410–1415.

  13. 13.

    Dariush, B. (2005) Analysis and simulation of an exoskeleton controller that accommodates static and reactive loads. In: Proceedings - IEEE International Conference on Robotics and Automation. pp 2350–2355.

  14. 14.

    Van Dijk, W., & Van Der Kooij, H. (2011). Hekman E A passive exoskeleton with artificial tendons: Design and experimental evaluation. IEEE International Conference on Rehabilitation Robotics.

    Article  PubMed  Google Scholar 

  15. 15.

    Collo, A., Bonnet, V., Venture, G. (2016) A quasi-passive lower limb exoskeleton for partial body weight support. In: Proceedings of the IEEE RAS and EMBS International Conference on Biomedical Robotics and Biomechatronics. (pp 643–648). doi:

  16. 16.

    Shi, D., Zhang, W., Zhang, W., & Ding, X. (2019). A review on lower limb rehabilitation exoskeleton robots. Chin J Mech Eng, 32(1), 74.

    CAS  Article  Google Scholar 

  17. 17.

    Sado, F., Yap, H. J., Ghazilla, R. A. R., & Ahmad, N. (2018). Exoskeleton robot control for synchronous walking assistance in repetitive manual handling works based on dual unscented kalman filter. PLoS One.

    Article  PubMed  PubMed Central  Google Scholar 

  18. 18.

    Song, G., Huang, R., Cheng, H., Chen, Q. (2019) Learning coupled parameters with continuous domains for human-powered lower exoskeleton. In: ICARM 2018 - 2018 3rd International Conference on Advanced Robotics and Mechatronics. (pp 189–194)

  19. 19.

    Xi, R., Zhu, Z., Du, F., Yang, M., Wang, X., Wu, Q. (2016) Design concept of the quasi-passive energy-efficient power-assisted lower-limb exoskeleton based on the theory of passive dynamic walking. In: 2016 23rd International Conference on Mechatronics and Machine Vision in Practice (M2VIP), (pp 1–5).

  20. 20.

    Aoustin, Y., & Formalskii, A. M. (2018). Walking of biped with passive exoskeleton: Evaluation of energy consumption. Multibody System Dynamics, 43(1), 71–96.

    Article  Google Scholar 

  21. 21.

    Lim, I.-s, Kwon, O., & Park, J. H. (2014). Gait optimization of biped robots based on human motion analysis. Robotics and Autonomous Systems, 62(2), 229–240.

    Article  Google Scholar 

  22. 22.

    Chen, Q., Cheng, H., Yue, C., Huang, R., & Guo, H. (2018). Dynamic balance gait for walking assistance exoskeleton. Applied Bionics and Biomechanics, 2018, 7847014.

    Article  PubMed  PubMed Central  Google Scholar 

  23. 23.

    Ghan, J., Steger, R., & Kazerooni, H. (2006). Control and system identification for the Berkeley lower extremity exoskeleton (BLEEX). Advanced Robotics, 20(9), 989–1014.

    Article  Google Scholar 

  24. 24.

    Yagn, N. (1890) Apparatus for facilitating walking, running, and jumping. U.S.A. Patent 440684, 1890/11/18

  25. 25.

    Reed, M.K. LIFESUIT exoskeleton gives the gift of walking so they shall walk. In: Proceedings of the 4th IEEE Global Humanitarian Technology Conference, GHTC 2014, 2014. pp 382–385. doi:

Download references


This manuscript was edited by Wallace Academic Editing.


This research is support by the National Chung-Shan Institute of Science and Technology under Grant No: NCSIST-401-V101(109) and in part by the Ministry of Science and Technology, Taiwan under Grant No: 108-2221-E-002-149-MY3.

Author information



Corresponding author

Correspondence to Jia-Yush Yen.

Ethics declarations

Conflict of interest

The authors report no conflict of interest.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Lee, CY., Lan, S.C., Lin, JJ. et al. Realization of Natural Human Motion on a 3D Biped Robot For Studying the Exoskeleton Effective. J. Med. Biol. Eng. (2021).

Download citation


  • Biped simulation
  • Human motion implementation
  • Exoskeleton effectiveness

JEL classification

  • C63