Alexander Schepelmann

The Robotics Institute

Carnegie Mellon University

Current Projects

Design and Control of an Antagonistically Actuated Robotic Leg that Matches Human Performance

Muscle-reflex based neuromuscular locomotion models present a promising alternative to existing control approaches of powered, segmented robotic legs. A legged robotic platform that matches human performance is required to test the full ability of such controllers. The Robotic Neuromuscular Leg 1 (RNL1) represents the initial development steps of a robotic gait testbed that can implement and test neuromuscular-inspired control ideas on robotic hardware. The robot can deliver fast motions that characterize human locomotion and can generate antagonistic co-contraction seen in neuromuscular systems. RNL1 can reliably generate human-like leg motions with high positional accuracy for joint speeds up to 190rpm, approximately 90% of a similarly sized human's maximum knee joint velocity. Video of the leg is available here.

L: Robotic Neuromuscular Leg (RNL). M-Top: Close-up of RNL1's series elastic actuator (SEA). M-Bottom: Close-up of the robot's compliant knee. R: Experimental trace of executed joint trajectory and corresponding SEA torques.  Red=Desired. Blue=Average. Green=Std. dev.

References

A. Schepelmann, M. D. Taylor, H. Geyer. Development of a Testbed for Robotic Neuromuscular Controllers. In Proceedings of the 2012 Robotics: Science and Systems Conference (2012 RSS), Sydney, Australia, July 2012. (.pdf)

Compact Nonlinear Springs with User Defined Torque-Deflection Profiles for Series Elastic Actuators

Series elastic actuators (SEAs) often use linear springs in their drivetrains, which requires design compromises between torque resolution and actuation bandwidth. Nonlinear springs (NLSs) with variable stiffness overcome this limitation. However current NLS designs rely on off-the-shelf metal springs, which increases their overall size, making them difficult to implement in compact and existing SEAs. This work presents an optimization-based synthesis method for NLSs that are compact and encode a user defined torque-deflection profile using elastic elements with an arbitrary stiffness profile. The presented prototype uses rubber as its elastic element, resulting in a compact design that generates the desired torque profile. Ongoing work concerns mitigating NLS hysteresis during spring downstroke through careful rubber selection and rubber state estimation using observers.

L: Nonlinear Spring (NLS) concept. M-Top: CAD renderings of an NLS cam that encodes an exponentially stiffening torque-deflection profile and the rubber elastic element. M-Bottom: Experimental testbed used to evaluate NLS components. R: Desired vs. Measured NLS profile.  Dots: Desired torque profile. Solid isocontours: Average torque profile for 15 deflections.

References

A. Schepelmann, K. A. Geberth, H. Geyer. Compact Nonlinear Springs with User Defined Torque-Deflection Profiles for Series Elastic Actuators. In Proceedings of the 2014 IEEE International Conference on Robotics and Automation (2014 IEEE ICRA), Hong Kong, China, May 2014. (.pdf)

Previous Projects

Real-Time Drivable Terrain Identification via HSI Color and Visual Texture

As part of my Master's Thesis at Case Western Reserve University, I worked on "CWRU Cutter" (pronounced "crew cutter"), an autonomous lawnmower developed for outdoor power equipment manufacturer MTD Products Inc. My research utilized computer vision to identify drivable terrain in front of the robot based on HSI color and edge-based visual texture in real-time. This information was used for reactive obstacle avoidance. The ultimate goal of my research was to develop an economically feasible, robust alternative to the first prototype's "Light Detection and Ranging (LIDAR)" unit for use on future commercial versions of the robot.

Since 2008, the robot has been entered into the Institute of Navigation's annual Robotic Lawnmower Competition. In 2009 CWRU Cutter won 1st place utilizing a combination of LIDAR and computer vision. In 2010, CWRU Cutter again won 1st place relying solely on computer vision.

L: CWRU Cutter. R:Obstacle identification results.

Selected References

A. Schepelmann, R. Hudson, F. Merat, R. D. Quinn. Visual Segmentation of Lawn Grass for a Mobile Robotic Lawnmower. In Proceedings of the 2010 IEEE/RSJ International Conference on Intelligent Robots and Systems (2010 IEEE/RSJ IROS), Taipei, Taiwan, October 2010. (.pdf)

K. A. Daltorio, A. D. Rolin, J. A. Beno, B. E. Hughes, A. Schepelmann, J. Green, M. S. Branicky, R. D. Quinn. An Obstacle-Edging Reflex for an Autonomous Lawnmower. In Proceedings of the 2010 IEEE/ION Position Location and Navigation Symposium (2010 ION/IEEE PLANS), Indian Wells, CA, May 2010. (.pdf)

A. Schepelmann, H. Snow, B. E. Hughes, J. Green, F. Merat, R. D. Quinn. Vision-Based Obstacle Detection and Avoidance for the CWRU Cutter Autonomous Lawnmower. In Proceedings of the 2009 IEEE International Conference on Technologies for Practical Robot Applications (2009 IEEE TePRA), Woburn, MA, November 2009. (.pdf)

Trackball Design and Prototyping for Blaberus Discoidalis Locomotion Study

As part of the Biorobots Team Research group, I designed and prototyped a novel 3 degree-of-freedom (DOF) stationary trackball for Roy E. Ritzmann of Case Western Reserve University's Department of Biology. Dr. Ritzmann uses the device to study Blaberus discoidalis cockroach locomotion and gait types via high-speed video analysis. The trackball incorporates hydrodynamic lubrication principles to allow a tethered cockroach to translate and rotate in place while being filmed from multiple angles simultaneously. The prototype replaced Dr. Ritzmann's previously used 1 DOF oil film setup, which only allowed a tethered cockroach to translate.