Carnegie Mellon University
The
Evaluation of Decentralized Reactive Swing-Leg Controllers on Powered Robotic Legs

Alexander Schepelmann
doctoral dissertation, tech. report CMU-RI-TR-16-02, Robotics Institute, Carnegie Mellon University, February, 2016


Download
  • Adobe portable document format (pdf) (5MB)
Copyright notice: This material is presented to ensure timely dissemination of scholarly and technical work. Copyright and all rights therein are retained by authors or by other copyright holders. All persons copying this information are expected to adhere to the terms and constraints invoked by each author's copyright. These works may not be reposted without the explicit permission of the copyright holder.

Abstract
We present work to transfer decentralized neuromuscular control strategies of human locomotion to powered segmented robotic legs. State-of-the-art robotic locomotion control approaches, like centralized planning and tracking in fully robotic systems and predefined motion pattern replay in prosthetic systems, do not enable the dynamism and reactiveness of able-bodied humans. Animals largely realize dexterous segmented leg performance with leg-encoded biomechanics and local feedback controls that bypass central processing. A decentralized neuromuscular controller was recently developed that enables robust locomotion for a simulated multi-segmented planar humanoid. A portion of this controller was used in an active ankle-foot prosthesis to modulate ankle torque during stance, enabling level and inclined ground walking. While these results suggest that the neuromuscular controller is a promising alternative control method for both fully robotic systems and powered prostheses, it is unclear if the controller can be transferred to multi-segmented robotic legs. The goal of this thesis is to investigate the feasibility of controlling a multi-segmented robotic leg with the proposed neuromuscular control approach, which may enable robots and powered prostheses to react to locomotion disturbances dynamically and in a human-like way. Specifically, work in this thesis investigates two hypotheses. Hypothesis one posits that the proposed decentralized swing-leg controllers enable more robust foot placements into ground targets than state-of-the-art impedance controls. Hypothesis two posits that neuromuscular swing-leg control enables more human-like motion than state-of-the-art impedance control.

To transfer neuromuscular controls to powered segmented robotic legs, we use a model-based design approach. The initial transfer is focused on neuromuscular swing-leg controls, important for maintaining dynamic stability of both fully robotic systems and powered prostheses in the presence of unexpected locomotion disturbances, such as trips and pushes. We first present the design of RNL, a three segment, cable-driven, antagonistically actuated robotic leg with joint compliance. The robot's size, weight, and actuation capabilities correspond to dynamically scaled human values. Next, a high-fidelity simulation of the robot is created to investigate the feasibility of transferring neuromuscular controls, pre-tune hardware gains via optimization, and serve as a benchmark for hardware experiments. An idealized version of the swing-leg controller with mono-articular actuation, as well as the neuromuscular interpretation of this controller with multi-articular actuation is then transferred to RNL and evaluated with foot placement experiments. The results suggest that the proposed swing-leg controllers can accurately regulate foot placement of robotic legs during undisturbed and disturbed motions. Compared to impedance control, the proposed controls achieve foot placements over a range of ground targets with a single set of gains, which make them attractive candidates for regulating the motion of legged robots and prostheses in the real-world. Furthermore, the ankle trajectory traced out by the robot under neuromuscular control is more human-like than the trajectories traced out under the proposed idealized control and impedance control. %although discrepancies exist between human motions and those executed by the robot. In parallel to this control transfer, a synthesis method for compact nonlinear springs with user-defined torque deflection profiles that use rubber as their compliant element is presented to explore methods for improving RNL's series elastic actuators.

In parallel to this control transfer, a synthesis method for creating compact nonlinear springs with user-defined torque-deflection profiles is presented to explore methods for improving RNL's series elastic actuators. The springs use rubber as their elastic element, which, while enabling a compact spring design, introduce viscoelastic behavior in the spring that needs to be accounted for with additional control. To accurately estimate force developed in the rubber, an empirically characterized constitutive rubber model is developed and integrated into the series elastic actuator controller used by the RNL test platforms. Benchtop experiments show that in conjunction with an observer, the nonlinear spring prototype achieves desired behavior at actuation frequencies up to 2 Hz, after which spring behavior degrades due to rubber hysteresis. These results show that while the presented methodology is capable of realizing compact nonlinear springs, careful rubber selection that mitigates viscoelastic behavior is necessary during the spring design process.

Keywords
Locomotion, Neuromuscular Control, Series Elastic Actuators, Nonlinear Springs

Notes

Text Reference
Alexander Schepelmann , "Evaluation of Decentralized Reactive Swing-Leg Controllers on Powered Robotic Legs," doctoral dissertation, tech. report CMU-RI-TR-16-02, Robotics Institute, Carnegie Mellon University, February, 2016

BibTeX Reference
@phdthesis{Schepelmann__2016_8055,
   author = "Alexander {Schepelmann }",
   title = "Evaluation of Decentralized Reactive Swing-Leg Controllers on Powered Robotic Legs",
   booktitle = "",
   school = "Robotics Institute, Carnegie Mellon University",
   month = "February",
   year = "2016",
   number= "CMU-RI-TR-16-02",
   address= "Pittsburgh, PA",
}