Embedded Files
5) Continuous and Dynamically Equilibrated One-Legged Running based on Virtual Admittance Control: Experiments on a Stiff-by-Nature Monopod:
What are the limitations of so called conventional legged robot technology (electrical actuators with harmonic gears)? Can we spice-up the things using sensory feedback? Can these robots exhibit agile locomotion characteristics instead of hanging around the laboratory. My answer to last 2 question is "Yes!" In this project, I worked on a classical one-legged robot with no passive compliance. It has a 3-DoF leg, actuated via AC servos with harmonic gears. It has no passive compliant element.
In this project, I developed a force feedback control method to ensure dynamic balance during running. It introduces virtual spring-damper admittance couple in each joint to reduce joint torque error. The same algorithm also appeared to be useful within the presence of various disturbances. It also enabled robot to perform monopedal running locomotion on various situations; such as, running on inclined platforms and 1 [cm] obstacles. Up to my knowledge, this project reports the first monopedal running over challenging environments in which a stiff-by-nature robot is used.
Related Publications: [IC1] Barkan Ugurlu, Takao Kawasaki, Michihiro Kawanishi, and Tatsuo Narikiyo, "Continuous and Dynamically Equilibrated One-Legged Running Experiments: Motion Generation and Indirect Force Feedback Control", in Proc. of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Algarve, Portugal, 2012. A journal paper is currently under preparation.
This project is carried out in Toyota Technological Institute (TTI), Nagoya, Japan. Video is a bit long, thus, feel free to travel between the scenes.
4) Continuous Hopping Locomotion Generation for Passive Compliant Bipeds via Exploiting the Base Resonance Frequency:
CoMan is a passively compliant bipedal robot actuated via Series Elastic Actutators (SEA) and built in Italian Institute of Technology (IIT), as a part of EU FP7 project AMARSi. It has initially built as a lower-body humanoid robot, but the folks in IIT are currently improving the existing mechatronic hardware and developing an upper body. Their ongoing research efforts on CoMan development can be viewed in their website.
Obviously, the inherent passively compliance enables this robot to achieve exhibit efficient hopping locomotion characteristics. To this end, I primarily determined the resonance frequencies given the dynamic model of CoMan. In addition to model-based approach, an additional system identification procedure is carried out to crosscheck computed resonance frequencies. Upon determining the resonance frequency, the vertical CoM trajectory is analytically derived; so that spring deflections, and therefore, elastic energy storage/release cycles are maximized. Horizontal motion is produced via ZMP approach. Carrying out the aforementioned procedures, CoMan was able to demonstrate successful hopping locomotion cycles which are repetitive and dynamically-balanced.
Related Publications: [IC3] Barkan Ugurlu, Jody A. Saglia, Nikos G. Tsagarakis, and Darwin G. Caldwell, Hopping at the Resonance Frequency: A Trajectory Generation Technique for Bipedal Robots with Elastic Joints, in Proc. of the IEEE International Conference on Robotics and Automation (ICRA), St. Paul, US, 2012, pp. 1436-1443. A journal paper is also currently under review.
Remarks: Back then, CoMan has a small mechanical design fault that limited its ankle joint range. Therefore, the robot was not able to perform hopping when the spring deflections were disabled. Keeping this in mind, this work reports one of the first experimental results where a bipedal robot is enabled to perform hopping through the utilization of passive compliance/base resonance frequency, as it cannot otherwise. This work is completed in collaboration with my former colleagues, Nikos G. Tsagarakis, Jody A. Saglia.
3) Bipedal Walking Energy Minimization for a Passively Compliant Robot: A Reinforcement Learning-based Approach with Evolving Policy Parameterization:
To what extent contemporary machine learning algorithms can be implemented to real-time humanoid locomotion? Should we start from the scratch, or should we combine machine learning with existing robotics approach? Can such methods fit in well in locomotion generation scenarios? I and my former colleague Petar blew our heads off to find answers to these questions while on our way to Nashville to attend IEEE Humanoids 2010. From Genoa, it was a 17 hours journey (changing planes 2 times) which gave us lots! of time to discuss, discuss, discuss more, settle down, start up the conversation, yet discuss more and finally converge to a solution to inject his AI skills in a bipedal locomotion generator.
The aim of this project is to minimize electrical energy (not mechanical energy!) while CoMan is walking. Biological structures, for instance us, humans, make use of the muscle passive compliance to manage elastic potential energy. By periodically storing/releasing elastic potential energy, the cost of transportation is observed to be reduced. In classical humanoid walking, the CoM (Center of Mass) height is constant. Since CoMan is equipped with springs in its joint, we generated a varying vertical (z-axis) CoM motion based on a reinforcement learning algorithm with evolving policy parameterization. x-axis and y-axis CoM trajectories are derived via ZMP concept, upon the existing vertical CoM trajectory. After 180 rollouts, CoMan learned to walk efficiently by exploiting the passive compliance in its joints.
Related Publications: [IC6] Petar Kormushev, Barkan Ugurlu, Sylvain Calinon, Nikos G. Tsagarakis, and Darwin G. Caldwell, Bipedal Walking Energy Minimization by Reinforcement Learning with Evolving Policy Parameterization, in Proc. of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), San Francisco, US, 2011, pp. 318-324.
[IC4] Petar Kormushev, Sylvain Calinon, Darwin G. Caldwell, and Barkan Ugurlu, Challenges for the Policy Representation when Applying Reinforcement Learning in Robotics, in Proc. of the IEEE International Joint Conference on Neural Networks (IJCNN), Brisbane, Australia, 2012, pp. 1-8.
A journal paper is also on the way.
Remarks: Up to our knowledge, this project reports the very first experimentally demonstrated bipedal walking energy minimization through the utilization of passive elements. This work is completed in collaboration with my former colleagues, Petar Kormushev, Nikos G. Tsagarakis.
2) Real-time Dynamic Walking Pattern Generation and Yaw Moment Compensation for the Passive Compliant Robot CoMan:
As I joined Italian Institute of Technology in 2010, I primarily composed a real-time dynamic walking pattern generator for CoMan. Since the great majority of AMARSi participants are specialized in computer and information science, a user-friendly and easy-to-implement walking generator was required. In doing so, we aimed to enable our partners in AMARSi project to conduct experiments using different motion parameters (time profile, lateral sway, stride length, max. swing foot height, etc.) without having the worry of dynamical balance. Therefore, the dynamic walking pattern generator appeared to operate in real-time and allowed users to alter motion parameters in online.
On top of this walking generator, I also implemented yaw moment compensation algorithm via the utilization of waist rotation. The main strategy in this method is to rotate the upper body in a way to exert a secondary moment that counteracts to the factors which create the undesired yaw moment. The technique allowed us to obtain 61% decrease in undesired yaw moment and 82% decrease in yaw axis deviation during walking.
Related Publications: [J2] Barkan Ugurlu, Jody A. Saglia, Nikos G. Tsagarakis, and Darwin G. Caldwell, Yaw Moment Compensation for Bipedal Robots via Intrinsic Angular Momentum Constraint, International Journal of Humanoid Robotics, Accepted, 2012, in print.
[IC7] Barkan Ugurlu, Nikos G. Tsagarakis, Emmanouil Spyrakos-Papastravridis, and Darwin G. Caldwell, Compliant Joint Modification and Real-Time Dynamic Walking Implementation on Bipedal Robot cCub, in Proc. of the IEEE International Conference on Mechatronics (ICM), Istanbul, Turkey, 2011, pp. 833-838.
Remarks: This work is completed in collaboration with my former colleagues, Nikos G. Tsagarakis, Jody A. Saglia. The following video only demonstrates the first baby steps of CoMan (formerly called cCub). Back then, CoMan had a small mechanical design fault that limited its ankle joint range. Therefore, the stride length had to be kept smaller. I guess it is Italy's first dynamically walking humanoid, operated in real-time. (Please correct me if I'm wrong.)
1) Eulerian ZMP Resolution: A Bipedal Trajectory Generator based on Combining Inertial Forces and Intrinsic Angular Momentum Rate Changes:
So-called conventional bipedal trajectory approach treats the whole robot as a point mass with zero rotational inertia. While useful in its own right, such a simplified model may not represent certain portion of robot dynamics. In other words, point-mass models cannot characterize the rotational inertia, a crucial element in representing humanoid dynamics and bipedal locomotion behavior.
To remedy this, I proposed a model in which the actual robot is considered as a 3-D composite rigid body with variant rotational inertia. Its CoM is assumed to be oriented in an identical manner to the composite body. In order to mimic the physical robot, actual composite rigid body inertia algorithm is utilized in this model, so that the rotational inertia is updated in each control cycle. Such a model allowed us to combine inertial forces and intrinsic angular momentum rate change terms in ZMP equations using spherical coordinates.
I ran bipedal walking and running simulations on ROCOS simulator. Moreover, real-time dynamic walking experiments are also conducted on the bipedal robot MARI-3. Several results show that the walking generator shows superior performance over the conventional point-mass models as it is based on a relatively more precise modeling approach.
Related Publications: [J1] Barkan Ugurlu, and Atsuo Kawamura, Bipedal Trajectory Generation based on Combining Inertial Forces and Intrinsic Angular Momentum Rate Changes: Eulerian ZMP Resolution, IEEE Transactions on Robotics, vol. 28, no. 6, December 2012, in print. [pdf]
[IC8] Barkan Ugurlu, and Atsuo Kawamura, Bipedal Walking Trajectory Generation based on ZMP and Euler's Equations of Motion,in Proc. of the IEEE International Conference on Humanoid Robots (Humanoids), Nashville, US, 2010, pp. 468-473.
[IC9] Barkan Ugurlu, and Atsuo Kawamura, Eulerian ZMP Resolution based Bipedal Walking: Discussions on the Intrinsic Angular Momentum Rate Change about Center of Mass, in Proc. of the IEEE International Conference on Robotics and Automation (ICRA), Anchorage, US, 2010, pp. 4218-4223.
Remarks: This work is completed during my Ph.D. studies under the supervision of Atsuo Kawamura, in Yokohama National University. Bipedal running is only simulated due to MARI-3's limited hardware specifications.
Google Sites
Report abuse