Integrating Passive Dynamic Wobbling with Leg …posa/DynamicWalking2020/614...Leg extension is a...

1
Integrating Passive Dynamic Wobbling with Leg Extension to Produce Stable Gaits in a Two-Actuator Bipedal Robot Sharfin Islam, Kamal Carter, Ryan St. Pierre, Sarah Bergbreiter, Aaron M. Johnson Mechanical Engineering Department, Carnegie Mellon University, Pittsburgh, PA, USA Email: [email protected], [email protected], [email protected] I. SUMMARY The long term goal of this project is to achieve bipedal walking at the small scales – the size of a Lego mini-figure (4cm). At this small scale, the number and power of the actuators will be extremely limited. This necessitates a walking gait that takes advantage of passive dynamics with only limited excitation from the actuators. In order to design these walking patterns and test the placement and power requirements on the actuators, in this abstract we present the design and control of a prototype that is 12 times larger, with 17cm legs. The robot, shown in Fig. 1, incorporates a single leg extension actuator per leg and leverages passive dynamic wobbling. With the current design, several different sinusoidal control trajectories were tested. The most successful being a 120° and 180° offset between the legs. Although walking is possible with this design and different trajectories, we found that the phasing of the wobble in the frontal plane and the leg extension were not what is predicted from a compas gait model. II. DESIGN Leg extension is a common actuation scheme for walking models [1,2] and in bipedal robot designs [3] (though in this case supported by a boom). For this project, with a target robot size of 4cm, our robot will not be supported by a boom arm and cannot fit multiple actuators. Therefore, we will use leg extension in combination with passive dynamics, as described by McGeer in [4], to compensate for the lack of hip actuation and achieve a stable gait without a fixed support. Refining the design and control using the prototype presented in this abstract will give us insight as to the viability of this approach at smaller scales. The current prototype, pictured in Fig. 1, is a 3D printed assembly with 7 components. For each leg, there is a motor housing, foot, and linkage. Additionally there is a torso about which the legs rotate. On the torso, the OpenCM9 controller and 7.5V Lithium Ion battery are mounted. Finally, there are two tabs on the torso that limit the swing angle to 20° in either direction of the center line. The linkages are attached to the ankle of the foot and driven by Dynamixel XL-320 servos. The feet are designed to be curved according to McGeer’s passive dynamics and are sloped downward to offset the center of mass and center of pressure. This offset gives the robot forward momentum once the legs clear the ground. III. RESULTS AND DISCUSSION With the current design, the robot is successfully able to walk with several different sinusoidal trajectories. With a 120° Fig. 1: (Left) Isometric View of the Robot from Back, (Right) Sagittal View of Robot showing Forward Lean offset, the gait is stable but the swing leg never goes past the stance leg. This gait was also surprising in that the push off was at the start of stance. Most walking models describe a push off at the end of stance after the swing leg is planted. With a 180° offset, the gait is stable and is more similar to other traditional walking patterns. This trajectory results in a push off at the end of stance after the swing leg touches down. However, this trajectory creates an inconsistent gait where the pitch and yaw from the wobble is sometimes perturbed by the leg extension. Although both of these trajectories result in a walking robot, they both need tuning in the design and control to successfully integrate leg extension and passive walking dynamics. In future work, a simulation model will be created to experiment with different aspects of the design such as: limit angle of the swing, offset difference between center of mass and center of pressure, and mass distribution. Additionally, different control methods will be tested such as integrating an IMU to control the extension based on the pitch of the robot. REFERENCES [1] M. Anand, J. Seipel, and S. Rietdyk, “A modelling approach to the dynamics of gait initiation,” Journal of the Royal Society Interface, vol. 14, no. 128, 2017. [2] S. Miyakoshi and G. Cheng, “Examining human walking characteristics with a telescopic compass-like biped walker model,” in IEEE Interna- tional Conference on Systems, Man and Cybernetics, vol. 2, November 2004, pp. 1538–1543. [3] A. M. Boudali, F. H. Kong et al., “Design and modeling of an open plat- form for dynamic walking research,” 2017 IEEE International Conference on Mechatronics (ICM), 2017. [4] T. McGeer and L. H. Palmer, “Wobbling, toppling, and forces of contact,” American Journal of Physics, vol. 57, no. 12, pp. 1089–1098, 1989.

Transcript of Integrating Passive Dynamic Wobbling with Leg …posa/DynamicWalking2020/614...Leg extension is a...

Page 1: Integrating Passive Dynamic Wobbling with Leg …posa/DynamicWalking2020/614...Leg extension is a common actuation scheme for walking models [1,2] and in bipedal robot designs [3]

Integrating Passive Dynamic Wobbling with Leg Extensionto Produce Stable Gaits in a Two-Actuator Bipedal Robot

Sharfin Islam, Kamal Carter, Ryan St. Pierre, Sarah Bergbreiter, Aaron M. JohnsonMechanical Engineering Department, Carnegie Mellon University, Pittsburgh, PA, USA

Email: [email protected], [email protected], [email protected]

I. SUMMARY

The long term goal of this project is to achieve bipedalwalking at the small scales – the size of a Lego mini-figure(4cm). At this small scale, the number and power of theactuators will be extremely limited. This necessitates a walkinggait that takes advantage of passive dynamics with only limitedexcitation from the actuators. In order to design these walkingpatterns and test the placement and power requirements on theactuators, in this abstract we present the design and controlof a prototype that is 12 times larger, with 17cm legs. Therobot, shown in Fig. 1, incorporates a single leg extensionactuator per leg and leverages passive dynamic wobbling.With the current design, several different sinusoidal controltrajectories were tested. The most successful being a 120° and180° offset between the legs. Although walking is possiblewith this design and different trajectories, we found that thephasing of the wobble in the frontal plane and the leg extensionwere not what is predicted from a compas gait model.

II. DESIGN

Leg extension is a common actuation scheme for walkingmodels [1,2] and in bipedal robot designs [3] (though in thiscase supported by a boom). For this project, with a target robotsize of 4cm, our robot will not be supported by a boom armand cannot fit multiple actuators. Therefore, we will use legextension in combination with passive dynamics, as describedby McGeer in [4], to compensate for the lack of hip actuationand achieve a stable gait without a fixed support. Refiningthe design and control using the prototype presented in thisabstract will give us insight as to the viability of this approachat smaller scales.

The current prototype, pictured in Fig. 1, is a 3D printedassembly with 7 components. For each leg, there is a motorhousing, foot, and linkage. Additionally there is a torso aboutwhich the legs rotate. On the torso, the OpenCM9 controllerand 7.5V Lithium Ion battery are mounted. Finally, there aretwo tabs on the torso that limit the swing angle to 20° in eitherdirection of the center line. The linkages are attached to theankle of the foot and driven by Dynamixel XL-320 servos. Thefeet are designed to be curved according to McGeer’s passivedynamics and are sloped downward to offset the center of massand center of pressure. This offset gives the robot forwardmomentum once the legs clear the ground.

III. RESULTS AND DISCUSSION

With the current design, the robot is successfully able towalk with several different sinusoidal trajectories. With a 120°

Fig. 1: (Left) Isometric View of the Robot from Back, (Right)Sagittal View of Robot showing Forward Lean

offset, the gait is stable but the swing leg never goes past thestance leg. This gait was also surprising in that the push offwas at the start of stance. Most walking models describe apush off at the end of stance after the swing leg is planted.With a 180° offset, the gait is stable and is more similar toother traditional walking patterns. This trajectory results in apush off at the end of stance after the swing leg touches down.However, this trajectory creates an inconsistent gait where thepitch and yaw from the wobble is sometimes perturbed by theleg extension.

Although both of these trajectories result in a walking robot,they both need tuning in the design and control to successfullyintegrate leg extension and passive walking dynamics. Infuture work, a simulation model will be created to experimentwith different aspects of the design such as: limit angle of theswing, offset difference between center of mass and center ofpressure, and mass distribution. Additionally, different controlmethods will be tested such as integrating an IMU to controlthe extension based on the pitch of the robot.

REFERENCES

[1] M. Anand, J. Seipel, and S. Rietdyk, “A modelling approach to thedynamics of gait initiation,” Journal of the Royal Society Interface,vol. 14, no. 128, 2017.

[2] S. Miyakoshi and G. Cheng, “Examining human walking characteristicswith a telescopic compass-like biped walker model,” in IEEE Interna-tional Conference on Systems, Man and Cybernetics, vol. 2, November2004, pp. 1538–1543.

[3] A. M. Boudali, F. H. Kong et al., “Design and modeling of an open plat-form for dynamic walking research,” 2017 IEEE International Conferenceon Mechatronics (ICM), 2017.

[4] T. McGeer and L. H. Palmer, “Wobbling, toppling, and forces of contact,”American Journal of Physics, vol. 57, no. 12, pp. 1089–1098, 1989.