Barkan Ugurlu

Bipedal walking energy minimization by reinforcement learning with evolving policy parameterization

We present a learning-based approach for minimizing the electric energy consumption during walking of a passively-compliant bipedal robot. The energy consumption is reduced by learning a varying-height center-of-mass trajectory which uses efficiently the robots passive compliance. To do this, we propose a reinforcement learning method which evolves the policy parameterization dynamically during the learning process and thus manages to find better policies faster than by using fixed parameterization. The method is first tested on a function approximation task, and then applied to the humanoid robot COMAN where it achieves significant energy reduction.

YAW MOMENT COMPENSATION FOR BIPEDAL ROBOTS VIA INTRINSIC ANGULAR MOMENTUM CONSTRAINT

This paper is aimed at describing a technique to compensate undesired yaw moment, which is inevitably induced about the support foot during single support phases while a bipedal robot is in motion. 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 moment. In order to compute the yaw moment by considering all the factors, we utilized Eulerian ZMP Resolution, as it is capable of characterizing the robots rotational inertia, a crucial component of its dynamics. In doing so, intrinsic angular momentum rate changes are smoothly included in yaw moment equations. Applying the proposed technique, we conducted several bipedal walking experiments using the actual bipedal robot CoMan. As the result, we obtained 61% decrease in undesired yaw moment and 82% regulation in yaw-axis deviation, which satisfactorily verify the efficiency of the proposed approach, in comparison to off-the-shelf techniques.

A framework for sensorless torque estimation and control in wearable exoskeletons

This paper is aimed at describing a framework to implement sensorless torque estimation and control in wearable exoskeletons, for the purpose of handling power augmentation tasks. The proposed method relies on accurately identifying and compensating the joint-level disturbance torques caused by stiction, viscous friction, and gravitational loads. Utilizing off-the-shelf techniques, the characteristics of these disturbances are primarily identified. Subsequently, additional torque inputs are superimposed to the system via feedforward loops in a way to counteract to these disturbances. Having compensated frictional and gravitational loads acting on the actuation module; we are able to estimate the external torque exerted at each joint by using disturbance observers. In this manner, torque control is enabled without any requirement of built-in torque sensing units. In order to validate the proposed framework, we conducted weight lifting and upholding experiments on able-bodied human subjects with and without wearing the upper extremity of exoskeleton suit. Comparison of EMG and IEMG signals acquired in two cases indicates that the exoskeleton system provides sufficient power augmentation reliably. In conclusion, the proposed method is validated to be efficient and it can be potentially used for rehabilitation, training and power augmentation tasks.

Dynamic trot-walking with the hydraulic quadruped robot — HyQ: Analytical trajectory generation and active compliance control

This paper presents a trajectory generator and an active compliance control scheme, unified in a framework to synthesize dynamic, feasible and compliant trot-walking locomotion cycles for a stiff-by-nature hydraulically actuated quadruped robot. At the outset, a CoP-based trajectory generator that is constructed using an analytical solution is implemented to obtain feasible and dynamically balanced motion references in a systematic manner. Initial conditions are uniquely determined for symmetrical motion patterns, enforcing that trajectories are seamlessly connected both in position, velocity and acceleration levels, regardless of the given support phase. The active compliance controller, used simultaneously, is responsible for sufficient joint position/force regulation. An admittance block is utilized to compute joint displacements that correspond to joint force errors. In addition to position feedback, these joint displacements are inserted to the position control loop as a secondary feedback term. In doing so, active compliance control is achieved, while the position/force trade-off is modulated via the virtual admittance parameters. Various trot-walking experiments are conducted with the proposed framework using HyQ, a ~ 75kg hydraulically actuated quadruped robot. We present results of repetitive, continuous, and dynamically equilibrated trot-walking locomotion cycles, both on level surface and uneven surface walking experiments.

ZMP-Based Online Jumping Pattern Generation for a One-Legged Robot

This paper is aimed at presenting a method to generate online jumping patterns, which can be applied to one-legged jumping robots and optionally to humanoid robots. Our proposed method is based on ensuring the overall dynamic balance through the complete jumping cycle. To be able to reach this goal, we discretized the zero moment point equation in polar coordinates so that we are able to include angular momentum information in a natural way. Thus, undesired torso angle fluctuation is expected to be more restrainable compared to other methods in which angular momentum information is ignored or zero referenced. Moreover, we unified support and flight phases in terms of motion generation. Having obtained successful simulation results and vertical jumping experiments in our previous work, we conducted forward jumping experiments. As the result, we obtained successful and repetitive jumping cycles, which satisfactorily verify the proposed method.

Yaw moment compensation of biped fast walking using 3D inverted pendulum

The restriction factors for fast biped walking are the problem of ZMP and the influence of the yaw moment around support leg. In this paper, we use 3D inverted pendulum model as the walking trajectory to resolve the problem of ZMP and improved the trajectory of double support phase to enable biped fast and stable walking. Then this paper proposes the compensation by rotating the waist joint, based on the swing leg trajectory to reduce the yaw moment. 40% decrease of the yaw moment and the improvement of walking stability proved by the 1.2[km/h] walking simulation. We succeeded 1.7[km/h] walking simulation as highest speed so for. In the experiment, we succeeded 0.35[km/h] walking at this moment.

Compliant joint modification and real-time dynamic walking implementation on bipedal robot cCub

Conventionally, humanoid robots consist of non-backdrivable stiff joint structures with high gain controllers, aiming at high precision. Although this feature enabled researchers to represent outstanding engineering prototypes, the resulting large mechanical impedance output makes these robots inherently unsafe when interacting with humans. Moreover, the same issue also limits their abilities to safely interact with the environment and reduces their energy efficiency. In order to cope with these problems, we developed a compliant robot called cCub, as a part of the European project AMARSi (Adaptive Modular Architectures for Rich Motor Skills). Having completed the design and realization of the lower body of the cCub robot, we developed an integrated framework, including the compliant experimental platform, a simulation environment and a walking pattern generator. Furthermore, we also conducted real-time dynamic walking experiments on cCub demonstrating that dynamically equilibrated walking cycles can be executed in this new highly passive compliant humanoid platform. These preliminary results validate the feasibility of the walking generator and the motion capability of the compliant robot.

Actively-compliant locomotion control on rough terrain: Cyclic jumping and trotting experiments on a stiff-by-nature quadruped

This paper is authored to describe a control framework that is designated for realizing cyclic, actively-compliant and dynamically-balanced jumping and trotting quadruped locomotion over rough terrain. In order to succeed in exhibiting such locomotion abilities, two controllers are synthesized: i) Active Compliance Control via force feedback, ii) Angular Momentum Control via gyro sensing. The first controller computes the joint displacements that are associated with ground reaction force errors, using Jacobian transpose and admittance blocks. Together with position constraints, these joint displacements are simultaneously fed-back to local servo controllers; allowing the robot to perform the given locomotion task in an actively-compliant manner. The second controller, in the meantime, evaluates gyro sensor information to calculate the required compensation torque about center of mass, which is necessary to regulate upper torso rotational motion. Afterwards, it updates the orientation input in accordance with this compensation torque. Using the proposed framework, the overall control performance is tested via cyclic jumping and trotting motion experiments, conducted over rough terrain with a stiff-by-nature quadruped robot. Results turn out to be positive; the robot demonstrates successful jumping and trotting cycles in a repetitive, actively-compliant and dynamically-balanced fashion.

Hopping at the resonance frequency: A trajectory generation technique for bipedal robots with elastic joints

It is known that bipedal robots with passive compliant structures have obvious advantages over stiff robots, as they are able to handle the potential energy management. Therefore, this paper is aimed at presenting a jumping pattern generation method that takes advantage of this property via the utilization of the base resonance frequency, which is of special importance. To begin with, the resonance frequency is determined through a system identification procedure on our actual robot. Consequentially, the vertical component of the CoM is generated via a periodic function in which the resonance frequency is employed. The horizontal component of the CoM is obtained using the ZMP criterion to guarantee the dynamic balance. Having obtained the necessary elements of the CoM trajectory within an analytical manner, joint motions are computed with the help of translational and angular momenta constraints. In order to validate the method, two legged jumping experiments are conducted on our actual compliant robot. In conclusion, we observed repetitive, continuous, and dynamically equilibrated jumping cycles with feasible landing phases.

Eulerian ZMP resolution based bipedal walking: Discussions on the intrinsic angular momentum rate change about center of mass

This paper is aimed at implementing Eulerian ZMP Resolution method to bipedal walking pattern generation. The main strategy in this method is to ensure the dynamic balance by generating feasible ZMP-based CoM trajectories. For this purpose, we employ ZMP equations in spherical coordinates, so that the intrinsic angular momentum rate change about center of mass is included explicitly in a natural way. This fact results in two merits: 1) Undesired torso angle fluctuation and body twists are expected to be more restrainable comparing to other methods in which intrinsic angular momentum information is ignored or zero-referenced. 2) The interference between motions in sagittal and lateral planes can be extracted. In this article, we mainly investigate the first merit and briefly discuss about the second merit. Applying the aforementioned technique, Eulerian ZMP Resolution, we simulated bipedal walking on a 3-D dynamic simulator. Secondarily, we conducted bipedal walking experiments on the actual bipedal robot. In conclusion, we obtained dynamically equilibrated bipedal walking cycles, which satisfactorily verify the efficiency of Eulerian ZMP Resolution technique over conventional methods.